Paper deep dive

Reference-Free Post-Training of Open Large Language Models for Multilingual Machine Translation

Chris Han, Pengzhi Gao, Pei Fu, Jian Luan

Year: 2026Venue: arXiv preprintArea: cs.CLType: PreprintEmbeddings: 205

Intelligence

Status: succeeded | Model: Gemma-4-26B-A4B | Prompt: intel-v1 | Confidence: 90%

Last extracted: 8/16/2026, 3:26:37 AM

Summary

This paper introduces MiLMMT-46-v1.0, a multilingual machine translation model series (1B, 4B, 12B parameters) covering 46 languages. The authors apply reference-free post-training using Group Relative Policy Optimization (GRPO) with a reward based on averaged quality estimation scores (XCOMET and COMETKiwi) gated by language identification. To mitigate reward-induced drift, they linearly interpolate supervised fine-tuning (SFT) and reinforcement learning (RL) checkpoints. The resulting models outperform strong open baselines (Seed-X, HY-MT2, TranslateGemma) and achieve leading reference-free scores against proprietary systems (Google Translate, Gemini 3 Pro, GPT-5). The study also investigates On-Policy Distillation (OPD) as an alternative, finding it reaches but does not surpass the quality frontier of RL with checkpoint interpolation.

Entities (15)

MiLMMT-46-v1.0 · model · 98%GRPO · algorithm · 95%MiLMMT-46-v0.1 · model · 95%COMETKiwi · metricmodel · 90%Gemini-3-pro · model · 90%Google Translate · model · 90%GPT-5 · model · 90%HY-MT2 · model · 90%On-Policy Distillation · method · 90%Seed-X · model · 90%TranslateGemma · model · 90%XCOMET · metricmodel · 90%FLORES · dataset · 85%OpenLID-v3 · model · 85%WMT24++ · dataset · 85%

Relation Signals (13)

MiLMMT-46-v1.0 → derivedfrom → MiLMMT-46-v0.1

confidence 95% · Starting from the supervised-finetuned MiLMMT-46-v0.1 models, we apply Group Relative Policy Optimization (GRPO)... to obtain MiLMMT-46-v1.0.

MiLMMT-46-v1.0 → usesmethod → GRPO

confidence 95% · Starting from the supervised-finetuned MiLMMT-46-v0.1 models, we apply Group Relative Policy Optimization (GRPO)

MiLMMT-46-v1.0 → achievesleadingscoreagainst → Google Translate

confidence 90% · achieve leading reference-free scores against evaluated proprietary systems such as Google Translate

MiLMMT-46-v1.0 → achievesleadingscoreagainst → GPT-5

confidence 90% · achieve leading reference-free scores against evaluated proprietary systems such as ... GPT-5.

MiLMMT-46-v1.0 → achievesleadingscoreagainst → Gemini-3-pro

confidence 90% · achieve leading reference-free scores against evaluated proprietary systems such as ... Gemini 3 Pro

MiLMMT-46-v1.0 → evaluatedon → WMT24++

confidence 90% · We evaluate multilingual translation performance on the FLORES+ and WMT24++ benchmarks.

MiLMMT-46-v1.0 → evaluatedon → FLORES

confidence 90% · We evaluate multilingual translation performance on the FLORES+ and WMT24++ benchmarks.

MiLMMT-46-v1.0 → →

Cypher Suggestions (0)

No Cypher suggestions yet.

Abstract

Abstract:We study reference-free post-training for multilingual machine translation with open large language models. Starting from the supervised-finetuned MiLMMT-46-v0.1 models, we apply Group Relative Policy Optimization (GRPO) with a reward that averages two reference-free quality estimation models and is gated by language identification. We then linearly interpolate the supervised fine-tuning (SFT) and reinforcement learning (RL) model checkpoints to obtain MiLMMT-46-v1.0. Across 46 languages, the resulting models consistently improve translation quality over their SFT counterparts, outperform strong recent open baselines, including Seed-X, HY-MT2, and TranslateGemma, and achieve leading reference-free scores against evaluated proprietary systems such as Google Translate, Gemini 3 Pro, and GPT-5. We further investigate on-policy distillation (OPD) and find that it reaches, but does not surpass, the quality frontier achieved by RL with checkpoint interpolation. We release the models and code to facilitate future research.

PDF

Open source PDF →Open local PDF →

Trouble viewing inline? Open PDF directly →

Full Text

204,118 characters extracted from source content.

Expand or collapse full text

Reference-Free Post-Training of Open Large Language Models for Multilingual Machine Translation Chris Han Pengzhi Gao Thanks: ˜Corresponding author: gaopengzhi@xiaomi.com. Pei Fu Jian Luan Affiliation: Xiaomi Inc., Beijing, China Abstract We study reference-free post-training for multilingual machine translation with open large language models. Starting from the supervised-finetuned MiLMMT-46-v0.1 models 18, we apply Group Relative Policy Optimization (GRPO) 19 with a reward that averages two reference-free quality estimation models and is gated by language identification. We then linearly interpolate the supervised fine-tuning (SFT) and reinforcement learning (RL) model checkpoints to obtain MiLMMT-46-v1.0. Across 46 languages, the resulting models consistently improve translation quality over their SFT counterparts, outperform strong recent open baselines including Seed-X 2, HY-MT2 22, and TranslateGemma 6, and achieve leading reference-free scores against evaluated proprietary systems such as Google Translate, Gemini 3 Pro, and GPT-5. We further investigate on-policy distillation (OPD) and find that it reaches, but does not surpass, the quality frontier achieved by RL with checkpoint interpolation. We release the models and code to facilitate future research.11 1 Models are released at https://huggingface.co/collections/xiaomi-research/milmmt-46. Codes are released at https://github.com/xiaomi-research/gemmax. 1 Introduction Open large language models (LLMs), including the Gemma and Qwen families 8; 15, have recently become strong foundations for multilingual machine translation (MT). Through continual pretraining and supervised finetuning (SFT), recent work has adapted these models into capable multilingual translation systems. In particular, the MiLMMT-46-v0.1 models 18 achieve strong performance across 4646 languages, providing a strong open starting point for further post-training. Despite these advances, further improving multilingual MT with supervised learning remains constrained by the scarcity and uneven coverage of high-quality parallel data, particularly for low-resource languages and non-English-centric translation directions. By contrast, source-side text is far more abundant and easier to collect across languages. This gap motivates reference-free post-training methods that can exploit source-side data without aligned target sentences. Reinforcement learning (RL) provides a natural framework for this setting because it can optimize sequence-level rewards on model-generated translations rather than token-level likelihood against references. Group Relative Policy Optimization (GRPO) 19 has recently been applied to multilingual translation with carefully designed rewards, as exemplified by Tower+ 16 and HY-MT2 22. Reference-free quality estimation (QE) models can score a source sentence and its candidate translation without a reference, but imperfect estimators may encourage reward hacking 12. A language-identification gate can suppress wrong-language outputs, while SFT–RL checkpoint interpolation can mitigate reward-induced drift and preserve behavior learned during SFT. These considerations motivate our combination of reference-free RL with checkpoint interpolation. Building on the MiLMMT-46-v0.1 models, we study this approach across three model scales and 4646 languages. Our experiments show that reference-free RL consistently improves learned quality metrics over SFT and compares favorably with strong recent open and proprietary systems, while checkpoint interpolation provides a controllable trade-off between reference-free quality and reference-based performance. We further investigate on-policy distillation (OPD) 1; 13 as an alternative post-training approach and find that it reaches, but does not surpass, the quality frontier achieved by RL with checkpoint interpolation. We release the resulting MiLMMT-46-v1.0 models and code to facilitate future research. 2 Methodology Starting from the supervised-finetuned MiLMMT-46-v0.1 models, we use Group Relative Policy Optimization (GRPO) to optimize a reference-free reward based on two quality estimation models and a language-identification gate. We then linearly interpolate the SFT and RL checkpoints to retain the behavior learned during supervised finetuning while incorporating the improvements from RL. Training objective. We adopt GRPO for the RL stage. For each source sentence x, we sample a group of G candidate translations y1,…,yG\y_1,…,y_G\ from the old policy πθold _ _old. Let RiR_i denote the reward assigned to candidate yiy_i, and let πref _ref denote the fixed SFT reference policy. We maximize the following objective: ⁡(θ)= (θ)= x,yi[1G∑i=1G1|yi|∑t=1|yi|min[ri,tA^i,t, _x,\,\y_i\ [ 1G _i=1^G 1|y_i| _t=1^|y_i| \ [r_i,t\, A_i,t, clip(ri,t,1−ϵ,1+ϵ)A^i,t]−βKL(πθ∥πref)], (r_i,t,1-ε,1+ε)\, A_i,t ]-β\,D_KL( _θ\| _ref) \ ], where ri,t(θ)=πθ(yi,t∣x,yi,<t)πθold(yi,t∣x,yi,<t)r_i,t(θ)= _θ(y_i,t x,y_i,<t) _ _old(y_i,t x,y_i,<t) is the token-level importance ratio, A^i,t A_i,t is the group-normalized advantage computed from the rewards Rii=1G\R_i\_i=1^G of the sampled translations, ϵε controls the clipping range, and β controls the Kullback–Leibler (KL) penalty that regularizes the RL policy toward the SFT reference policy. Reward design. Given a source sentence x, a candidate translation y, and the intended target language ℓ , we predict the language of y, denoted by ℓ^(y) (y), using OpenLID-v3 5. Let sXs_X and sKs_K denote the scores assigned to (x,y)(x,y) by XCOMET22 2 https://huggingface.co/Unbabel/XCOMET-XXL 9 and COMETKiwi33 3 https://huggingface.co/Unbabel/wmt23-cometkiwi-da-xxl 17, respectively. We define the reward as R⁡(x,y,ℓ)=12(sX+sK),ℓ^(y)=ℓ,0,otherwise.R(x,y, )= cases 12 (s_X+s_K ),& (y)= ,\\ 0,&otherwise. cases This language-identification gate prevents fluent but wrong-language outputs from receiving high QE rewards, mitigating a known form of reward hacking 12. RL data. We derive the RL dataset from the MiLMMT SFT data by retaining each example’s source sentence and translation direction while discarding its reference translation. This yields 263982263982 instances spanning 192192 translation directions. Because GRPO relies on within-group reward variation, rollout groups with nearly identical rewards provide little training signal. We therefore sample G translations for each instance using the RL rollout configuration, score them with our reward function, and retain instances whose group mean μx _x and standard deviation σx _x satisfy 0.30<μx<0.950.30< _x<0.95 and σx≥0.05 _x≥ 0.05. The mean bounds remove instances with very low or high average rewards, while the standard-deviation threshold removes groups with little reward variation. This leaves 3157231572 instances, randomly split into 3057230572 training and 10001000 validation instances. A detailed breakdown by translation direction is provided in Appendix A. SFT–RL checkpoint interpolation. Let θSFT _SFT denote the parameters of MiLMMT-46-v0.1 and θRL _RL the parameters obtained after RL. To reduce reward-induced drift while retaining the gains from RL, we linearly interpolate the two checkpoints: θα=αθSFT+(1−α)θRL,α∈[0,1]. _α=α _SFT+(1-α) _RL, α∈[0,1]. Here, α controls the contribution of the SFT checkpoint, while 1−α1-α controls that of the RL checkpoint. The interpolation requires no additional training. We refer to θRL _RL as MiLMMT-46-v0.1-RL and the selected interpolated checkpoint as MiLMMT-46-v1.0. 3 Experimental Setup Datasets. We conduct experiments on 4646 languages spanning a broad linguistic spectrum, with detailed language information summarized in Table 4. We evaluate multilingual translation performance on the FLORES+ 14 and WMT24++ 4 benchmarks. For WMT24++, we adopt the English source sentences and exclude those marked as low quality for reference-free evaluation. Baselines. We compare against strong proprietary systems, including Google Translate, Gemini 2.5/3 Pro, and GPT-5, as well as the large-scale multilingual NMT model NLLB 21. In addition, we compare against several strong open-source multilingual translation models: • Tower-Plus 16: Gemma2/Qwen2.5-based models for multilingual translation and general-purpose tasks across 2727 languages. • GemmaX2-28 3: Gemma2-based models designed for multilingual machine translation across 2828 languages. • Seed-X-Instruct/PPO 2: Mistral-based models trained with instruction finetuning and reinforcement learning for multilingual machine translation across 2828 languages. • Hunyuan-MT, HY-MT1.5, and HY-MT2 24; 23; 22: Hunyuan-based multilingual translation models across 3333 languages. • TranslateGemma 6: Gemma3-based models for high-quality translation across 5555 languages. Training and checkpoint interpolation configurations. We apply the same RL recipe at all three scales (11B, 44B, and 1212B), using the verl44 4 https://github.com/verl-project/verl framework 20 with vLLM-based 11 rollouts. For each RL instance, we sample G=8G=8 candidates and regularize the policy toward the SFT reference πref _ref with a low-variance KL penalty. After RL training, we interpolate the SFT and RL parameters to construct the final checkpoint. For the MiLMMT-46-v1.0 models, we set the interpolation coefficient to α=0.5α=0.5 at all three scales. We explore other values of α in the interpolation analysis. A complete list of hyper-parameters is provided in Appendix C. Evaluation. We use XCOMET and COMETKiwi as reference-free evaluators on both the FLORES+ and WMT24++ benchmarks. Both are 1010B-parameter models with high correlation with human judgments 7. On FLORES+, we additionally report reference-based spBLEU and XCOMET scores. For each baseline, we evaluate on the subset of languages shared with our 4646-language model. Translations are generated using greedy decoding. Model WMT24++ FLORES+ en → x en → x x → en zh → x x → zh QE Ref. QE Ref. QE Ref. QE Ref. QE 2121 languages Tower-Plus-2B 82.52/76.30 40.41/92.65 94.10/88.53 42.54/94.61 94.88/89.77 24.27/88.56 87.99/78.87 29.53/84.65 71.29/76.52 Tower-Plus-9B 86.80/80.76 43.33/95.09 95.93/91.05 45.32/96.03 95.84/90.44 27.59/92.65 90.73/80.68 33.25/88.09 73.95/78.59 Tower-Plus-72B 86.25/80.21 42.74/94.86 95.76/90.71 45.89/96.26 96.00/90.51 28.50/92.92 90.55/80.41 35.06/89.17 74.61/78.88 MiLMMT-46-1B-v1.0 83.28/76.69 36.87/92.02 93.88/88.03 41.10/94.45 95.15/89.73 22.94/87.49 87.57/78.29 28.20/85.39 74.03/77.78 MiLMMT-46-4B-v1.0 89.11/82.77 39.85/95.64 96.67/91.85 43.45/96.06 96.33/90.57 26.53/92.83 91.38/82.07 31.47/89.49 76.77/80.45 MiLMMT-46-12B-v1.0 90.77/84.55 40.66/96.60 97.39/92.88 44.59/96.55 96.53/90.77 28.03/94.43 92.28/83.14 32.19/90.79 78.01/81.43 2626 languages Seed-X-Instruct-7B 85.19/78.30 44.16/94.45 95.30/90.17 44.54/93.10 91.92/88.67 28.60/90.43 85.68/80.96 32.51/87.60 68.36/79.97 Seed-X-PPO-7B 86.18/79.99 45.48/95.02 95.95/91.10 44.12/95.92 95.73/90.51 29.65/92.80 89.32/83.88 29.20/88.86 68.90/80.88 MiLMMT-46-1B-v1.0 83.42/77.21 37.72/92.39 94.19/88.66 41.72/94.28 94.87/89.82 23.12/87.32 86.75/80.83 27.89/85.04 73.44/79.45 MiLMMT-46-4B-v1.0 89.19/83.25 40.50/95.87 96.86/92.40 44.30/95.94 96.06/90.68 26.72/92.75 90.63/84.79 31.36/89.32 76.30/82.31 MiLMMT-46-12B-v1.0 90.88/85.00 41.18/96.84 97.58/93.46 45.43/96.45 96.28/90.89 28.15/94.34 91.58/85.74 32.39/90.72 77.58/83.28 2828 languages GemmaX2-28-2B-v0.2 80.95/77.35 37.38/90.14 92.42/87.69 42.29/92.52 91.86/88.28 25.06/86.56 85.32/79.71 29.70/83.57 68.60/78.17 GemmaX2-28-9B-v0.2 84.19/80.28 40.04/92.66 94.32/89.56 44.53/94.02 92.88/88.81 27.67/90.27 87.94/81.75 32.39/86.42 70.41/79.78 MiLMMT-46-1B-v1.0 79.73/75.97 33.12/88.23 91.22/86.37 38.73/91.24 91.23/87.73 21.43/83.24 83.36/78.35 26.28/81.48 68.84/77.60 MiLMMT-46-4B-v1.0 86.13/81.81 36.16/93.19 95.05/90.30 42.05/93.86 93.09/88.75 25.01/90.03 88.47/82.29 30.11/86.55 71.54/80.32 MiLMMT-46-12B-v1.0 88.09/83.57 36.82/94.58 96.01/91.39 43.44/94.62 93.52/89.00 26.39/91.93 89.55/83.19 31.35/88.20 72.84/81.30 3131 languages HY-MT1.5-1.8B 83.88/77.38 24.81/88.51 91.80/85.31 25.25/88.51 87.51/84.01 17.60/82.79 84.93/74.74 22.61/82.58 70.06/77.06 Hunyuan-MT-7B 84.51/80.71 27.84/89.97 92.37/88.53 30.20/89.90 88.29/85.76 20.83/85.89 84.55/78.08 22.98/85.35 71.69/79.15 HY-MT1.5-7B 84.42/80.85 28.97/90.16 92.37/88.78 31.68/90.44 88.87/86.20 20.61/86.38 85.45/79.35 23.59/86.18 72.45/79.10 HY-MT2-1.8B 82.61/78.30 30.97/88.40 91.14/86.20 32.29/88.57 88.29/85.83 21.31/85.17 84.87/76.82 24.82/82.60 70.89/75.74 HY-MT2-7B 86.21/81.86 34.84/92.10 93.92/89.68 38.79/92.57 91.36/87.78 23.24/89.29 87.69/80.14 25.35/88.08 73.79/78.36 HY-MT2-30B-A3B 86.44/82.58 34.94/92.30 93.87/90.24 40.07/93.90 91.98/88.08 24.38/89.92 87.79/78.25 29.91/88.83 73.68/78.60 MiLMMT-46-1B-v1.0 78.95/75.94 32.54/87.82 90.76/86.40 37.95/91.06 90.80/87.37 21.89/83.36 83.74/77.14 26.64/81.85 69.44/76.82 MiLMMT-46-4B-v1.0 85.48/81.82 35.61/92.90 94.71/90.28 41.24/93.75 92.69/88.39 25.30/89.85 88.66/80.74 30.36/86.79 72.12/79.31 MiLMMT-46-12B-v1.0 87.50/83.59 36.27/94.33 95.73/91.35 42.68/94.58 93.18/88.65 26.61/91.81 89.76/81.83 31.44/88.38 73.35/80.30 4646 languages Google Translate 83.29/80.46 42.90/91.39 92.71/88.97 47.42/93.95 91.94/89.24 30.74/89.08 83.94/78.26 36.08/85.35 63.80/76.48 Gemini 3 Pro 85.03/81.70 42.42/94.17 95.13/91.20 46.44/94.87 92.50/89.35 29.90/92.22 88.49/81.66 33.81/87.45 68.48/78.49 Gemini 2.5 Pro 84.49/81.46 41.15/93.85 94.85/90.97 46.13/94.64 92.34/89.30 29.12/91.93 88.12/81.35 33.07/87.54 68.54/78.68 GPT-5 84.86/82.10 38.42/93.06 94.25/91.07 43.64/94.58 92.14/89.29 26.36/91.43 88.64/81.92 31.34/87.46 70.06/79.02 NLLB-54.5B – 38.05/86.86 88.51/83.17 43.23/90.55 88.32/87.16 25.17/82.37 80.61/75.64 20.72/65.31 47.49/60.65 TranslateGemma-4B 75.97/73.03 27.71/82.77 85.87/82.78 33.42/88.43 86.14/86.65 17.71/78.13 77.90/73.45 23.37/80.34 64.54/75.85 TranslateGemma-12B 84.16/81.63 31.05/91.20 93.63/90.32 35.45/91.98 89.30/88.01 21.56/87.25 85.56/81.38 26.75/85.95 68.32/79.50 TranslateGemma-27B 86.08/83.31 32.44/93.46 95.14/91.67 38.31/93.61 91.21/88.83 23.47/89.89 87.60/82.66 28.75/87.31 68.76/80.11 MiLMMT-46-1B-v0.1 76.14/73.08 35.07/86.35 89.33/85.11 40.61/90.68 90.00/88.06 22.12/80.82 81.06/75.85 26.94/79.80 64.61/75.32 MiLMMT-46-4B-v0.1 83.27/79.93 39.56/92.05 93.91/89.79 43.93/93.61 92.25/89.26 26.41/88.91 87.45/80.91 30.96/85.05 67.87/78.57 MiLMMT-46-12B-v0.1 85.09/81.75 41.24/93.46 94.88/90.88 45.45/94.37 92.76/89.52 28.27/91.00 88.60/82.05 32.44/86.65 68.95/79.57 MiLMMT-46-1B-v1.0 79.01/75.82 33.99/88.13 91.11/86.87 39.52/91.35 90.90/88.34 21.72/82.77 83.21/78.09 26.80/81.52 67.08/77.01 MiLMMT-46-4B-v1.0 85.86/82.20 37.28/93.29 95.09/91.07 42.68/93.94 92.74/89.38 25.33/89.83 88.45/82.29 30.56/86.60 69.93/79.80 MiLMMT-46-12B-v1.0 87.88/84.06 38.04/94.68 96.09/92.19 44.04/94.70 93.18/89.64 26.79/91.83 89.62/83.41 31.68/88.18 71.20/80.79 Table 1: Translation performance on WMT24++ and FLORES+. WMT24++ QE cells report reference-free XCOMET / COMETKiwi. For FLORES+, Ref. cells report spBLEU / reference-based XCOMET, while QE cells report reference-free XCOMET / COMETKiwi. The best and second-best scores for each metric within each language-coverage block are shown in bold and underlined, respectively. Detailed reference-free and reference-based results are provided in Appendix D.1 and Appendix D.2, respectively. 4 Experimental Results 4.1 Main Results Table 1 reports reference-free results on WMT24++ and both reference-free and reference-based results on FLORES+. Since the compared systems differ in language coverage, each block uses the language subset shared by its included systems. Effects of reference-free post-training. Within the full 4646-language block, MiLMMT-46-v1.0 consistently improves over its supervised-finetuned MiLMMT-46-v0.1 counterpart across all three model scales. Averaged over the three model scales, XCOMET and COMETKiwi scores on WMT24++ improve by 2.752.75 and 2.442.44 points, respectively. On FLORES+, averaged over the four direction groups and three model scales, reference-based XCOMET, reference-free XCOMET, and COMETKiwi improve by 1.171.17, 1.411.41, and 1.171.17 points, while spBLEU decreases by 1.211.21 points. These results show that reference-free post-training consistently improves learned quality metrics, with a modest reduction in lexical overlap. Since spBLEU measures lexical overlap with a single reference translation and can penalize valid alternative wording, we view its modest decrease as less informative than the consistent gains in learned quality metrics. Full 4646-language comparison. MiLMMT-46-12B-v1.0 achieves the highest reference-free XCOMET and COMETKiwi scores on WMT24++ and across all four FLORES+ direction groups, outperforming the evaluated proprietary systems, including Google Translate, Gemini 2.5/3 Pro, and GPT-5. On reference-based XCOMET, it ranks among the top two systems in three of the four FLORES+ direction groups. Among the smaller models, MiLMMT-46-1B-v1.0 outperforms TranslateGemma-4B on every reported WMT24++ and FLORES+ metric despite using only one quarter as many parameters. Although its spBLEU remains below Google Translate and Gemini 2.5/3 Pro, MiLMMT-46-12B-v1.0 surpasses GPT-5 in three of the four FLORES+ direction groups and achieves a higher average across the four groups (35.1435.14 vs. 34.9434.94). It also achieves a higher average spBLEU across the four groups than all evaluated external open-source baselines, including NLLB-54.5B and TranslateGemma-27B. Comparisons on shared language subsets. Across the 2121-, 2626-, 2828-, and 3131-language subsets, MiLMMT-46-12B-v1.0 ranks first in 1515 of the 1616 reference-based XCOMET comparisons across the four FLORES+ direction groups. It also achieves the best reference-free scores in nearly all comparisons against Tower-Plus, Seed-X, GemmaX2, and the HY-MT series. These results are not limited to the largest model. MiLMMT-46-4B-v1.0 remains competitive with substantially larger models, including Tower-Plus-72B and HY-MT2-30B-A3B. 4.2 Analysis Figure 1: Training and validation rewards during GRPO post-training at the 11B, 44B, and 1212B scales. Validation rewards are evaluated every 5050 training steps. Figure 2: Trade-off between spBLEU and reference-based XCOMET under SFT–RL checkpoint interpolation on FLORES+, averaged over four translation-direction groups: en → x, x → en, zh → x, and x → zh. Each colored curve corresponds to one model scale, and α denotes the SFT weight. The MiLMMT-46-v0.1-RL and MiLMMT-46-v0.1 checkpoints form the two endpoints of each curve. Circled points indicate the selected MiLMMT-46-v1.0 checkpoints at α=0.5α=0.5. External systems are included for comparison. Reference-free post-training gains persist across model scales. Figure 1 shows that training and validation rewards increase early and then stabilize at all three model scales. Comparing MiLMMT-46-v1.0 with its SFT counterpart v0.1 on WMT24++, reference-free XCOMET improves by 2.872.87, 2.592.59, and 2.792.79 points at the 11B, 44B, and 1212B scales, respectively. COMETKiwi shows the same trend. These results indicate that reference-free post-training remains effective across model scales, including on top of the strongest 1212B SFT model, rather than being limited to smaller models. Interpolation as a controllable quality trade-off. Figure 2 shows a consistent trade-off between spBLEU and reference-based XCOMET across all three model scales. As the SFT weight α increases, the interpolated checkpoints move toward the SFT endpoint: spBLEU increases monotonically, while reference-based XCOMET decreases. The selected α=0.5α=0.5 checkpoints provide a favorable balance along these curves. Relative to the RL endpoints, they recover 2.792.79, 3.703.70, and 4.214.21 spBLEU points at the 11B, 44B, and 1212B scales, respectively, while reducing reference-based XCOMET by only 0.530.53, 0.560.56, and 0.540.54 points. Increasing α beyond 0.50.5 yields diminishing returns in spBLEU. For the 1212B model, increasing α from 0.50.5 to 0.70.7 gains another 1.311.31 spBLEU points at a cost of 0.550.55 XCOMET, whereas increasing it from 0.70.7 to 0.90.9 gains only 0.370.37 spBLEU points at a further cost of 0.390.39 XCOMET. We therefore fix α=0.5α=0.5 across model scales as a simple operating point that recovers a substantial portion of the spBLEU lost during RL while retaining most of the improvement in reference-based XCOMET. 5 On-Policy Distillation Analysis Model WMT24++ FLORES+ 1B MiLMMT-46-1B-v1.0 79.01 30.51 / 85.94 OPD (init. v0.1) 77.67 30.47 / 85.22 RL + OPD (init. v0.1) 78.25 30.49 / 85.67 OPD (init. v1.0) 77.86 30.39 / 85.42 4B MiLMMT-46-4B-v1.0 85.86 33.96 / 90.91 OPD (init. v0.1) 85.42 34.17 / 90.82 RL + OPD (init. v0.1) 85.85 33.98 / 90.98 OPD (init. v1.0) 85.37 34.16 / 90.79 Table 2: On-policy distillation from MiLMMT-46-12B-v1.0 into 11B and 44B students, averaged over 4646 languages. The v1.0 baseline is obtained through RL and checkpoint interpolation, and parenthetical labels indicate the student initialization. WMT24++ reports reference-free XCOMET, while FLORES+ reports spBLEU and reference-based XCOMET. RL + OPD adopts λ=1λ=1. Full per-group results and results for all tested values of λ are provided in Appendix E. We study whether on-policy distillation (OPD) can transfer the gains of a post-trained teacher to smaller students without running reinforcement learning separately at each model scale. We use MiLMMT-46-12B-v1.0 as the teacher and train 11B and 44B students using the same reference-free RL data described in Section 2. The students generate their own rollouts, and the teacher provides a distillation signal at the student-induced states. Given a prompt x, a student rollout y∼πθ(⋅∣x)y _θ(· x), and a state st=(x,y<t)s_t=(x,y_<t), OPD trains the student by minimizing the expected token-level divergence between the student policy and a teacher policy: ℒOPD(θ)=x,y∼πθ[1|y|∑t=1|y|D(πθ(⋅∣st),πϕ(⋅∣st),yt)] _OPD(θ)=E_x,\,y _θ\! [ 1|y| _t=1^|y|D\! ( _θ(· s_t), _φ(· s_t),y_t ) ] where πϕ _φ denotes the teacher policy and D is a per-token divergence signal. This objective admits two implementations. The distributional form, GKD 1, defines D as a distribution-level KL and backpropagates through it directly. The policy-gradient form, PG-OPD 13, estimates the divergence from student-sampled tokens and optimizes the resulting signal as a reward through an RL update. We adopt PG-OPD as our primary implementation to study OPD and RL under a unified objective. Unless stated otherwise, OPD refers to the policy-gradient form. We report GKD only as a control experiment in Appendix E. Specifically, PG-OPD instantiates the per-token divergence for a student-sampled token yty_t as Dt=sg⁡(log⁡πθ(yt∣st)−log⁡πϕ(yt∣st)),D_t=sg\! ( _θ(y_t s_t)- _φ(y_t s_t) ), which is a single-sample estimator of the reverse KL divergence from the student policy to the teacher policy. Since ℒOPDL_OPD averages DtD_t over student-sampled tokens, minimizing it is equivalent to maximizing its negative. We therefore use rt=−Dtr_t=-D_t as the per-token reward and optimize it with the same GRPO objective described in Section 2. Here, sg⁡(⋅)sg(·) denotes the stop-gradient operator, which prevents gradients from flowing through the reward computation while preserving its dependence on the teacher policy. We denote the resulting policy-gradient distillation loss by ℒdistillL_distill and the GRPO loss from Section 2 by ℒpolicyL_policy. We compare three training variants that differ in their objectives and student initialization. OPD (init. v0.1) optimizes ℒdistillL_distill alone starting from MiLMMT-46-v0.1. RL+OPD (init. v0.1) combines the GRPO and distillation losses: ℒ=ℒpolicy+λℒdistillL=L_policy+λ\,L_distill where λ controls the distillation weight. OPD (init. v1.0) instead optimizes ℒdistillL_distill alone starting from the corresponding MiLMMT-46-v1.0 student. Transferring post-training gains through OPD. Table 2 compares the distilled students with the corresponding MiLMMT-46-v1.0 baselines. OPD (init. v0.1) closely matches the 44B baseline on FLORES+, achieving 34.1734.17 versus 33.9633.96 spBLEU and 90.8290.82 versus 90.9190.91 reference-based XCOMET, while the 11B student remains slightly behind in reference-based XCOMET (85.2285.22 versus 85.9485.94). Overall, OPD from the post-trained 1212B teacher approaches the performance of the corresponding MiLMMT-46-v1.0 models, but does not consistently match or surpass them. Distillation weight controls the quality trade-off. Varying λ in RL + OPD produces a trade-off similar to that obtained through SFT–RL checkpoint interpolation in Section 4.2. At 44B, reducing λ from 11 to 0.0010.001 increases reference-based XCOMET from 90.9890.98 to 91.4791.47 while decreasing spBLEU from 33.9833.98 to 30.2030.20, with intermediate values following the same trend. With λ=1λ=1, RL + OPD remains close to the corresponding MiLMMT-46-v1.0 baselines across both benchmarks and model scales. Initializing OPD from MiLMMT-46-v1.0 rather than MiLMMT-46-v0.1 likewise has little effect. Overall finding. Across objectives, distillation weights, and student initializations, OPD and RL + OPD remain close to the quality frontier established by RL and checkpoint interpolation but do not extend it. We therefore view OPD as a robustness finding and an alternative way to transfer post-training gains to smaller models, rather than an improvement over RL and interpolation. 6 Conclusion In this paper, we study reference-free post-training for multilingual machine translation with open large language models, analyzing GRPO, SFT–RL checkpoint interpolation, and on-policy distillation across the 11B, 44B, and 1212B model scales. Experiments covering 4646 languages on WMT24++ and FLORES+ show that language-gated quality estimation rewards consistently improve reference-free XCOMET and COMETKiwi, as well as reference-based XCOMET, although the resulting RL checkpoints exhibit lower spBLEU. Building on MiLMMT-46-v0.1, we develop MiLMMT-46-v1.0 by interpolating the SFT and RL checkpoints, recovering a substantial portion of the spBLEU lost during RL while retaining most of the gains in neural quality metrics. The resulting models compare favorably with strong recent open-source baselines, and MiLMMT-46-12B-v1.0 achieves leading reference-free results among the evaluated open and proprietary systems. Our on-policy distillation experiments further show that the gains of the post-trained 1212B teacher can be transferred to smaller students, although distillation does not improve on the trade-off achieved by RL and checkpoint interpolation. We hope our findings and released models support scalable and reference-free post-training for open multilingual translation systems. In future work, we will explore more robust and human-aligned reward models, extend reference-free post-training to multimodal multilingual translation, and study its behavior at larger model scales. References Agarwal et al. (2024) R. Agarwal, N. Vieillard, Y. Zhou, P. Stanczyk, S. Ramos Garea, M. Geist, and O. Bachem On-policy distillation of language models: learning from self-generated mistakes. In International Conference on Learning Representations, Vol. 2024, p. 21246–21263. Cited by: §1, §5. Cheng et al. (2025) S. Cheng, Y. Bao, Q. Cao, L. Huang, L. Kang, Z. Liu, Y. Lu, W. Zhu, J. Chen, Z. Huang, T. Li, Y. Li, H. Lin, S. Liu, N. Peng, S. She, L. Xu, N. Xu, S. Yang, R. Yu, Y. Yu, L. Zou, H. Li, L. Lu, Y. Wang, and Y. Wu Seed-x: building strong multilingual translation llm with 7b parameters. External Links: 2507.13618, Link Cited by: 3rd item, Abstract. Cui et al. (2025) M. Cui, P. Gao, W. Liu, J. Luan, and B. Wang Multilingual machine translation with open large language models at practical scale: an empirical study. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), L. Chiruzzo, A. Ritter, and L. Wang (Eds.), Albuquerque, New Mexico, p. 5420–5443. External Links: Link, Document, ISBN 979-8-89176-189-6 Cited by: 2nd item. Deutsch et al. (2025) D. Deutsch, E. Briakou, I. R. Caswell, M. Finkelstein, R. Galor, J. Juraska, G. Kovacs, A. Lui, R. Rei, J. Riesa, S. Rijhwani, P. Riley, E. Salesky, F. Trabelsi, S. Winkler, B. Zhang, and M. Freitag WMT24++: expanding the language coverage of WMT24 to 55 languages & dialects. In Findings of the Association for Computational Linguistics: ACL 2025, W. Che, J. Nabende, E. Shutova, and M. T. Pilehvar (Eds.), Vienna, Austria, p. 12257–12284. External Links: Link, Document, ISBN 979-8-89176-256-5 Cited by: §3. Fedorova et al. (2026) M. Fedorova, N. Arefyev, M. Buljan, J. Helcl, S. Oepen, E. Rønningstad, and Y. Scherrer OpenLID-v3: improving the precision of closely related language identification – an experience report. In Proceedings of the 13th Workshop on NLP for Similar Languages, Varieties and Dialects, Y. Scherrer, N. Aepli, V. Blaschke, T. Jauhiainen, N. Ljubešić, P. Nakov, J. Tiedemann, and M. Zampieri (Eds.), Rabat, Morocco, p. 275–292. External Links: Link, Document Cited by: §2. Finkelstein et al. (2026) M. Finkelstein, I. Caswell, T. Domhan, J. Peter, J. Juraska, P. Riley, D. Deutsch, C. Dilanni, C. Cherry, E. Briakou, E. Nielsen, J. Luo, K. Black, R. Mullins, S. Agrawal, W. Xu, E. Kats, S. Jaskiewicz, M. Freitag, and D. Vilar TranslateGemma technical report. External Links: 2601.09012, Link Cited by: 5th item, Abstract. Freitag et al. (2023) M. Freitag, N. Mathur, C. Lo, E. Avramidis, R. Rei, B. Thompson, T. Kocmi, F. Blain, D. Deutsch, C. Stewart, C. Zerva, S. Castilho, A. Lavie, and G. Foster Results of WMT23 metrics shared task: metrics might be guilty but references are not innocent. In Proceedings of the Eighth Conference on Machine Translation, P. Koehn, B. Haddow, T. Kocmi, and C. Monz (Eds.), Singapore, p. 578–628. External Links: Link, Document Cited by: §3. Gemma Team (2026) Gemma Team Gemma 4 technical report. External Links: 2607.02770, Link Cited by: §1. Guerreiro et al. (2024) N. M. Guerreiro, R. Rei, D. v. Stigt, L. Coheur, P. Colombo, and A. F. T. Martins XCOMET: transparent machine translation evaluation through fine-grained error detection. Transactions of the Association for Computational Linguistics 12, p. 979–995. External Links: Link, Document Cited by: §2. Joshi et al. (2020) P. Joshi, S. Santy, A. Budhiraja, K. Bali, and M. Choudhury The state and fate of linguistic diversity and inclusion in the NLP world. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, D. Jurafsky, J. Chai, N. Schluter, and J. Tetreault (Eds.), Online, p. 6282–6293. External Links: Link, Document Cited by: Table 4. Kwon et al. (2023) W. Kwon, Z. Li, S. Zhuang, Y. Sheng, L. Zheng, C. H. Yu, J. E. Gonzalez, H. Zhang, and I. Stoica Efficient memory management for large language model serving with pagedattention. In Proceedings of the ACM SIGOPS 29th Symposium on Operating Systems Principles, Cited by: §3. Liu et al. (2026) Y. Liu, S. Ouyang, H. R. Yatish, and L. Li Mending the holes: mitigating reward hacking in reinforcement learning for multilingual translation. In Findings of the Association for Computational Linguistics: ACL 2026, M. Liakata, V. P. Moreira, J. Zhang, and D. Jurgens (Eds.), San Diego, California, United States, p. 33699–33723. External Links: Link, Document Cited by: §1, §2. Lu and Lab (2025) K. Lu and T. M. Lab On-policy distillation. Thinking Machines Lab: Connectionism. Note: https://thinkingmachines.ai/blog/on-policy-distillation External Links: Document Cited by: §1, §5. NLLB Team et al. (2024) NLLB Team, M. R. Costa-jussà, J. Cross, O. Çelebi, M. Elbayad, K. Heafield, K. Heffernan, E. Kalbassi, J. Lam, D. Licht, J. Maillard, A. Sun, S. Wang, G. Wenzek, A. Youngblood, B. Akula, L. Barrault, G. M. Gonzalez, P. Hansanti, J. Hoffman, S. Jarrett, K. R. Sadagopan, D. Rowe, S. Spruit, C. Tran, P. Andrews, N. F. Ayan, S. Bhosale, S. Edunov, A. Fan, C. Gao, V. Goswami, F. Guzmán, P. Koehn, A. Mourachko, C. Ropers, S. Saleem, H. Schwenk, and J. Wang Scaling neural machine translation to 200 languages. Nature 630 (8018), p. 841–846. External Links: ISSN 1476-4687, Document, Link Cited by: §3. Qwen Team (2025) Qwen Team Qwen3 technical report. External Links: 2505.09388, Link Cited by: §1. Rei et al. (2025) R. Rei, N. M. Guerreiro, J. Pombal, J. Alves, P. Teixeirinha, A. Farajian, and A. F. T. Martins Tower+: bridging generality and translation specialization in multilingual llms. External Links: 2506.17080, Link Cited by: §1, 1st item. Rei et al. (2023) R. Rei, N. M. Guerreiro, J. Pombal, D. van Stigt, M. Treviso, L. Coheur, J. G. C. de Souza, and A. F. T. Martins Scaling up CometKiwi: unbabel-IST 2023 submission for the quality estimation shared task. In Proceedings of the Eighth Conference on Machine Translation, P. Koehn, B. Haddow, T. Kocmi, and C. Monz (Eds.), Singapore, p. 841–848. External Links: Link, Document Cited by: §2. Shang et al. (2026) Y. Shang, P. Gao, W. Liu, J. Luan, and J. Su Scaling model and data for multilingual machine translation with open large language models. External Links: 2602.11961, Link Cited by: Appendix A, §1, Abstract. Shao et al. (2024) Z. Shao, P. Wang, Q. Zhu, R. Xu, J. Song, X. Bi, H. Zhang, M. Zhang, Y. K. Li, Y. Wu, and D. Guo DeepSeekMath: pushing the limits of mathematical reasoning in open language models. External Links: 2402.03300, Link Cited by: §1, Abstract. Sheng et al. (2025) G. Sheng, C. Zhang, Z. Ye, X. Wu, W. Zhang, R. Zhang, Y. Peng, H. Lin, and C. Wu HybridFlow: a flexible and efficient rlhf framework. In Proceedings of the Twentieth European Conference on Computer Systems, p. 1279–1297. Cited by: §3. Team et al. (2022) N. Team, M. R. Costa-jussà, J. Cross, O. Çelebi, M. Elbayad, K. Heafield, K. Heffernan, E. Kalbassi, J. Lam, D. Licht, J. Maillard, A. Sun, S. Wang, G. Wenzek, A. Youngblood, B. Akula, L. Barrault, G. M. Gonzalez, P. Hansanti, J. Hoffman, S. Jarrett, K. R. Sadagopan, D. Rowe, S. Spruit, C. Tran, P. Andrews, N. F. Ayan, S. Bhosale, S. Edunov, A. Fan, C. Gao, V. Goswami, F. Guzmán, P. Koehn, A. Mourachko, C. Ropers, S. Saleem, H. Schwenk, and J. Wang No language left behind: scaling human-centered machine translation. External Links: 2207.04672, Link Cited by: §3. Zheng et al. (2026) M. Zheng, Z. Li, T. Chen, B. Lv, M. Sun, M. Song, J. Song, H. Huang, D. Wu, H. Wang, Y. Song, Y. Chen, and G. Zhang Hy-mt2: a family of fast, efficient and powerful multilingual translation models in the wild. External Links: 2605.22064, Link Cited by: §1, 4th item, Abstract. Zheng et al. (2025a) M. Zheng, Z. Li, T. Chen, M. Song, and D. Wang HY-mt1.5 technical report. External Links: 2512.24092, Link Cited by: 4th item. Zheng et al. (2025b) M. Zheng, Z. Li, B. Qu, M. Song, Y. Du, M. Sun, and D. Wang Hunyuan-mt technical report. External Links: 2509.05209, Link Cited by: 4th item. Appendix A Distribution of RL Training Data After effective-data selection, the retained pool comprises 3157231572 source sentences (3057230572 for training and 10001000 held out for validation) across 192192 translation directions. Table 3 reports the number of sentences per direction, following the language-code convention of 18. Direction Train Val Direction Train Val Direction Train Val en → ru 2205 73 cs → ta 71 2 hu → es 32 1 en → cs 1776 68 pl → yue 68 4 zhs → hi 32 1 en → de 1479 47 hu → th 68 1 ur → ru 31 2 en → ja 1327 43 ta → sl 63 0 ro → pt 31 1 en → bn 1219 38 bn → ar 62 0 uz → k 31 0 ja → en 1189 47 ru → tr 60 0 nb → fa 30 2 en → zhs 775 18 my → en 57 3 bg → sk 30 1 bn → en 726 19 hi → ta 56 1 cs → el 30 1 km → en 689 24 th → tr 55 2 it → th 30 1 ru → en 664 25 el → uz 53 4 lo → en 29 2 en → tr 644 22 fr → az 53 4 sl → da 29 2 de → en 620 17 sk → lo 53 1 da → ms 29 1 en → fr 612 15 el → km 53 0 fr → nb 29 1 en → he 598 18 fi → id 52 3 k → nl 29 1 tr → en 510 22 az → tl 52 1 ca → bg 28 0 fr → de 463 17 hi → ms 51 0 bg → ro 27 1 en → pt 446 11 hr → tr 51 0 my → zht 27 0 de → fr 441 13 my → ur 50 3 ca → ru 25 0 en → uz 437 23 ur → bg 50 2 es → nl 25 0 cs → en 389 16 ur → hr 50 1 ms → sv 25 0 en → km 362 8 el → fi 49 4 fa → de 22 2 en → az 359 10 fi → ur 48 3 nb → bg 22 2 en → my 358 12 sk → ar 48 3 ru → es 22 1 en → it 336 12 it → tr 48 1 es → fr 22 0 en → tl 315 8 hr → zht 47 1 k → ca 21 1 en → yue 306 11 bg → fi 47 0 pl → es 21 0 en → ur 300 13 ur → nl 46 0 es → bn 20 2 fi → en 300 12 az → km 45 3 el → ro 20 0 en → ta 294 5 tr → ca 45 2 fr → es 20 0 en → es 279 6 zht → k 45 1 pt → ms 20 0 en → hi 278 7 ca → hu 45 0 it → ru 19 0 en → nl 272 9 sv → tl 44 3 ar → en 18 0 en → hu 271 6 ca → lo 44 1 k → fr 16 0 en → sk 266 11 ta → vi 44 1 th → en 16 0 en → hr 265 9 pl → ja 44 0 uz → es 16 0 en → ro 258 7 hr → ms 43 3 yue → en 16 0 en → sl 257 15 sl → km 43 3 es → sv 14 1 ta → en 246 7 ko → ru 43 2 ur → en 14 1 k → en 226 11 nl → he 43 1 de → ru 12 0 en → ar 224 7 bn → it 42 2 hu → en 12 0 en → lo 223 7 zht → fr 41 4 sl → ru 11 2 en → zht 223 3 fr → zhs 41 2 ca → es 11 0 en → pl 212 5 sl → pt 41 2 ko → en 11 0 he → en 196 5 ta → sv 41 2 sk → en 11 0 en → fa 194 7 tr → ro 41 2 sl → en 11 0 en → fi 192 6 el → hr 40 2 fa → en 10 0 en → ca 182 1 zhs → lo 40 1 hr → en 10 0 zhs → en 173 7 tl → km 39 4 ca → en 9 0 en → bg 168 3 az → id 39 0 ms → en 9 0 en → th 167 5 my → nl 39 0 tl → en 9 0 en → el 164 7 fa → ms 38 1 bg → en 8 1 ro → en 152 5 uz → en 38 0 id → de 8 0 en → ms 148 9 ar → he 37 2 pt → it 8 0 fr → en 140 5 az → en 37 2 el → en 6 0 en → ko 128 3 de → hi 37 0 vi → en 6 0 en → id 122 4 ca → hi 36 2 nb → en 4 0 en → k 116 4 bn → zhs 36 0 da → en 3 0 pl → en 114 5 el → ca 35 3 it → es 3 0 en → vi 108 4 ru → pt 35 1 pt → en 3 0 hi → en 99 4 yue → da 35 0 sv → en 3 0 en → sv 96 1 az → da 34 1 zht → en 3 0 id → en 81 1 hi → id 34 1 it → en 2 0 en → da 79 3 fa → fr 33 2 nl → en 2 0 en → nb 79 3 nb → ro 33 0 es → en 1 0 Table 3: Number of instances for each translation direction in the RL training data, sorted by training count. All 192192 directions are shown (30,57230,572 training / 1,0001,000 validation in total). Appendix B Language Details Table 4 provides information on the 46 languages supported by our models. ISO Code Language Script Family Subgrouping Resource ar Arabic Arabic Afro-Asiatic Semitic High az Azerbaijani Latin Turkic Common Turkic Low bg Bulgarian Cyrillic Indo-European Balto-Slavic Mid bn Bengali Bengali Indo-European Indo-Aryan Mid ca Catalan Latin Indo-European Italic High cs Czech Latin Indo-European Balto-Slavic High da Danish Latin Indo-European Germanic Mid de German Latin Indo-European Germanic High el Greek Greek Indo-European Graeco-Phrygian Mid en English Latin Indo-European Germanic High es Spanish Latin Indo-European Italic High fa Persian Arabic Indo-European Iranian High fi Finnish Latin Uralic Finnic High fr French Latin Indo-European Italic High he Hebrew Hebrew Afro-Asiatic Semitic Mid hi Hindi Devanagari Indo-European Indo-Aryan High hr Croatian Latin Indo-European Balto-Slavic High hu Hungarian Latin Uralic - High id Indonesian Latin Austronesian Malayo-Polynesian Mid it Italian Latin Indo-European Italic High ja Japanese Japanese Japonic Japanesic High k Kazakh Cyrillic Turkic Common Turkic Mid km Khmer Khmer Austroasiatic Khmeric Low ko Korean Hangul Koreanic Korean High lo Lao Lao Tai-Kadai Kam-Tai Low ms Malay Latin Austronesian Malayo-Polynesian Mid my Burmese Myanmar Sino-Tibetan Burmo-Qiangic Low nb Norwegian Latin Indo-European Germanic Low nl Dutch Latin Indo-European Germanic High pl Polish Latin Indo-European Balto-Slavic High pt Portuguese Latin Indo-European Italic High ro Romanian Latin Indo-European Italic Mid ru Russian Cyrillic Indo-European Balto-Slavic High sk Slovak Latin Indo-European Balto-Slavic Mid sl Slovenian Latin Indo-European Balto-Slavic Mid sv Swedish Latin Indo-European Germanic High ta Tamil Tamil Dravidian South Dravidian Mid th Thai Thai Tai-Kadai Kam-Tai Mid tl Tagalog Latin Austronesian Malayo-Polynesian Mid tr Turkish Latin Turkic Common Turkic High ur Urdu Arabic Indo-European Indo-Aryan Mid uz Uzbek Latin Turkic Common Turkic Mid vi Vietnamese Latin Austroasiatic Vietic High yue Cantonese Han Sino-Tibetan Sinitic Low zhs Chinese (Simplified) Han Sino-Tibetan Sinitic High zht Chinese (Traditional) Han Sino-Tibetan Sinitic Low Table 4: 4646 languages supported by our model. The resource of each language is determined according to the taxonomy classes by 10. Appendix C Training Hyper-parameters Table 5 lists the complete set of GRPO hyper-parameters, shared across the 11B, 44B, and 1212B models. Hyper-parameter Value Optimization Actor learning rate 1e−61e-6 Batch size (prompts) 128128 PPO mini-batch size 128128 Total epochs 33 KL loss type low-variance KL KL coefficient β 1e−31e-3 Gradient checkpointing Enabled Rollout / sampling Inference backend vLLM Group size G 88 Temperature 0.80.8 Top-p 0.950.95 Max prompt length 40964096 tokens Max response length 40964096 tokens Reward (reference-free) Quality metrics XCOMET, COMETKiwi Language-matched reward 12(XCOMET+COMETKiwi) 12(XCOMET+COMETKiwi) Language-mismatch reward 00 Language identification OpenLID-v3 Table 5: GRPO training hyper-parameters, shared across all three model scales. Appendix D Detailed Evaluation Results D.1 Reference-Free Results Tables 6–11 report the per-direction reference-free results (XCOMET / COMETKiwi). Direction TranslateGemma-4B TranslateGemma-12B TranslateGemma-27B GemmaX2-28-2B GemmaX2-28-9B Hunyuan-MT-7B HY-MT1.5-1.8B HY-MT1.5-7B HY-MT2-1.8B HY-MT2-7B HY-MT2-30B-A3B Seed-X-Instruct-7B Seed-X-PPO-7B Tower-Plus-2B Tower-Plus-9B Tower-Plus-72B en → ar 82.93 / 78.21 86.54 / 81.66 87.12 / 83.01 77.96 / 74.33 82.34 / 78.80 88.87 / 82.85 85.12 / 78.51 88.98 / 82.66 83.57 / 78.39 87.70 / 82.24 88.27 / 82.49 80.33 / 75.78 81.80 / 77.34 - / - - / - - / - en → az 39.43 / 45.81 71.65 / 76.53 77.85 / 81.72 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → bg 83.19 / 77.11 88.78 / 83.66 89.77 / 84.98 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → bn 78.79 / 75.57 84.04 / 79.68 84.95 / 80.72 78.80 / 77.55 81.61 / 79.78 86.50 / 80.30 83.05 / 75.73 86.77 / 80.81 81.93 / 77.34 86.77 / 81.94 87.50 / 82.22 - / - - / - - / - - / - - / - en → ca 79.59 / 71.97 85.65 / 78.91 86.59 / 80.12 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → cs 80.07 / 75.12 86.54 / 81.74 87.55 / 83.54 78.78 / 71.86 83.84 / 77.78 88.05 / 82.53 82.61 / 75.88 87.58 / 82.28 82.60 / 74.60 87.49 / 80.68 88.50 / 82.13 83.43 / 76.69 84.35 / 77.67 76.91 / 69.83 83.61 / 77.82 83.06 / 76.78 en → da 90.04 / 81.51 92.74 / 85.16 93.64 / 86.29 - / - - / - - / - - / - - / - - / - - / - - / - 90.64 / 80.69 90.67 / 81.25 87.69 / 77.84 89.94 / 80.45 90.01 / 80.16 en → de 93.06 / 77.75 93.22 / 80.63 94.79 / 81.52 92.23 / 76.92 93.67 / 80.11 95.37 / 81.85 94.43 / 78.08 95.35 / 81.66 93.04 / 77.62 94.86 / 80.95 95.13 / 81.74 92.96 / 78.49 92.90 / 79.11 90.98 / 75.69 93.53 / 79.49 93.50 / 79.62 en → el 80.47 / 74.71 86.77 / 81.08 88.44 / 83.48 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → es 89.00 / 78.97 90.61 / 81.65 90.90 / 82.53 87.25 / 77.30 89.35 / 80.00 92.03 / 82.68 91.73 / 80.90 92.20 / 82.50 89.81 / 78.91 91.60 / 81.03 92.14 / 82.07 87.51 / 76.46 87.88 / 77.68 86.48 / 76.19 88.95 / 79.12 89.26 / 79.77 en → fa 79.77 / 77.50 85.90 / 83.49 87.56 / 84.94 78.73 / 78.25 83.25 / 82.22 86.53 / 82.36 82.33 / 77.46 86.45 / 82.69 81.30 / 78.27 86.63 / 83.59 87.53 / 84.72 - / - - / - - / - - / - - / - en → fi 83.44 / 80.94 88.96 / 86.47 90.49 / 88.28 - / - - / - - / - - / - - / - - / - - / - - / - 87.09 / 82.39 87.74 / 84.15 80.14 / 77.12 87.95 / 84.78 84.80 / 80.91 en → fr 84.52 / 78.40 87.37 / 81.47 87.44 / 82.54 83.77 / 76.44 86.09 / 79.03 88.73 / 82.09 87.31 / 78.83 88.97 / 82.00 86.28 / 78.09 87.67 / 79.95 89.02 / 81.71 85.68 / 77.51 85.42 / 78.28 81.46 / 74.36 86.28 / 79.80 86.85 / 79.88 en → he 76.75 / 71.40 84.05 / 79.29 86.31 / 81.14 78.41 / 73.43 82.23 / 77.52 78.48 / 70.66 76.35 / 68.93 78.34 / 71.17 76.90 / 70.31 84.75 / 78.57 85.67 / 79.95 - / - - / - - / - - / - - / - en → hi 75.17 / 68.43 82.54 / 72.27 83.76 / 73.75 73.22 / 69.70 76.77 / 71.42 86.01 / 70.11 82.27 / 66.61 86.03 / 71.64 80.16 / 70.42 84.96 / 73.68 85.67 / 74.11 - / - - / - 66.30 / 66.88 73.08 / 70.57 70.08 / 69.26 en → hr 80.35 / 75.45 87.39 / 82.84 89.60 / 85.46 - / - - / - - / - - / - - / - - / - - / - - / - 82.19 / 75.46 83.71 / 77.94 - / - - / - - / - en → hu 75.53 / 71.64 86.27 / 82.75 89.31 / 85.95 - / - - / - - / - - / - - / - - / - - / - - / - 86.25 / 81.26 87.24 / 83.06 80.16 / 76.36 86.62 / 83.10 83.46 / 79.47 en → id 88.84 / 81.48 91.54 / 84.58 91.59 / 84.77 87.36 / 80.40 88.77 / 81.96 92.94 / 85.37 91.85 / 83.07 92.80 / 85.53 90.54 / 82.87 92.53 / 84.87 92.81 / 85.16 85.14 / 77.06 85.79 / 78.69 - / - - / - - / - en → it 87.65 / 79.94 88.70 / 82.44 89.63 / 83.06 85.89 / 76.99 87.81 / 79.74 91.37 / 83.68 89.98 / 80.96 91.50 / 83.48 89.03 / 80.09 90.69 / 82.30 91.36 / 82.78 86.11 / 76.94 86.12 / 77.99 84.57 / 75.76 88.33 / 80.42 87.99 / 80.57 en → ja 83.02 / 84.36 87.23 / 87.40 88.47 / 88.05 84.31 / 85.03 86.62 / 86.65 90.91 / 88.98 88.71 / 86.96 90.83 / 88.90 87.95 / 86.93 90.69 / 88.78 90.92 / 88.80 83.15 / 82.85 85.19 / 84.73 83.05 / 84.49 86.60 / 86.88 87.39 / 87.18 en → k 39.80 / 55.35 64.89 / 79.54 73.00 / 84.14 - / - - / - 34.93 / 68.37 60.20 / 61.48 34.76 / 69.33 41.52 / 59.03 46.32 / 68.28 36.97 / 69.43 - / - - / - - / - - / - - / - en → km 37.64 / 46.67 64.89 / 77.61 70.23 / 81.16 71.41 / 79.63 75.22 / 82.12 68.36 / 78.57 70.81 / 78.65 67.70 / 78.73 75.00 / 81.98 79.01 / 84.43 78.29 / 84.62 - / - - / - - / - - / - - / - en → ko 85.97 / 82.51 89.88 / 85.85 90.64 / 86.69 84.02 / 82.15 88.07 / 85.49 92.21 / 87.24 89.95 / 84.97 91.77 / 87.06 88.47 / 84.94 91.96 / 87.14 91.74 / 87.23 79.96 / 77.72 83.58 / 81.42 81.83 / 80.54 86.78 / 84.44 87.09 / 84.32 en → lo 38.64 / 46.84 68.05 / 75.22 75.19 / 79.77 71.15 / 76.09 76.36 / 79.57 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → ms 83.00 / 76.45 88.75 / 82.31 89.23 / 82.52 85.13 / 78.27 86.58 / 79.42 91.58 / 83.08 90.82 / 80.83 91.44 / 83.35 89.17 / 81.03 90.72 / 82.57 91.10 / 83.08 81.27 / 73.07 82.98 / 75.48 - / - - / - - / - en → my 39.71 / 56.00 55.85 / 75.06 63.73 / 79.88 66.36 / 78.74 73.53 / 82.35 62.45 / 76.34 66.14 / 77.59 62.74 / 76.23 71.04 / 81.55 75.00 / 83.04 76.80 / 84.68 - / - - / - - / - - / - - / - en → nb 89.83 / 82.89 92.36 / 86.34 93.42 / 87.03 - / - - / - - / - - / - - / - - / - - / - - / - 86.75 / 76.16 88.51 / 79.61 86.80 / 78.31 87.08 / 78.40 86.53 / 77.89 en → nl 90.10 / 80.01 92.89 / 84.14 93.67 / 85.03 89.74 / 79.56 92.24 / 82.94 92.60 / 82.34 90.91 / 79.30 92.25 / 82.37 89.46 / 79.20 93.26 / 83.24 94.06 / 84.81 91.43 / 80.74 91.88 / 82.37 88.66 / 78.27 91.68 / 82.18 91.86 / 82.59 en → pl 85.33 / 76.77 89.51 / 82.21 90.29 / 83.63 83.41 / 74.66 87.14 / 79.11 86.22 / 78.20 84.10 / 74.69 86.31 / 78.11 83.38 / 73.85 89.28 / 80.77 91.35 / 82.83 87.45 / 78.50 88.00 / 79.62 81.70 / 72.82 88.18 / 80.01 87.22 / 78.84 en → pt 89.87 / 81.24 91.26 / 83.63 91.57 / 84.42 88.22 / 79.29 90.19 / 81.81 92.56 / 84.15 91.50 / 81.97 92.59 / 84.48 90.67 / 81.01 91.44 / 82.21 92.20 / 83.28 87.36 / 77.40 88.11 / 79.23 86.20 / 77.25 89.02 / 80.76 89.26 / 81.47 en → ro 83.72 / 82.33 88.16 / 87.34 89.52 / 88.09 - / - - / - - / - - / - - / - - / - - / - - / - 86.08 / 83.14 87.52 / 84.38 81.97 / 78.30 87.23 / 84.50 85.65 / 82.90 en → ru 85.17 / 79.73 88.25 / 82.81 88.87 / 83.85 81.14 / 75.02 84.86 / 79.51 90.02 / 83.33 86.06 / 78.89 90.01 / 83.52 84.86 / 78.41 88.70 / 82.25 89.49 / 83.18 84.78 / 78.99 85.10 / 80.49 80.04 / 75.08 85.31 / 80.36 85.65 / 80.70 en → sk 75.63 / 70.91 84.95 / 80.79 87.59 / 83.53 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → sl 70.34 / 64.91 82.87 / 78.64 86.33 / 82.73 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → sv 89.82 / 81.91 92.91 / 85.60 93.53 / 86.66 - / - - / - - / - - / - - / - - / - - / - - / - 90.55 / 80.97 90.41 / 82.06 87.83 / 79.23 90.98 / 83.43 89.99 / 81.62 en → ta 69.67 / 74.96 76.08 / 79.13 77.06 / 79.83 - / - - / - 70.30 / 72.31 72.58 / 70.75 70.97 / 72.75 72.35 / 75.43 78.78 / 80.50 79.27 / 81.05 - / - - / - - / - - / - - / - en → th 82.36 / 77.82 88.04 / 83.72 89.62 / 84.48 82.19 / 78.63 85.92 / 81.95 90.82 / 84.24 87.15 / 80.34 91.07 / 84.38 85.85 / 80.98 90.55 / 84.72 90.89 / 85.04 79.35 / 75.31 81.69 / 77.66 - / - - / - - / - en → tl 76.58 / 75.76 80.69 / 79.63 80.40 / 79.89 76.90 / 75.65 80.64 / 77.71 74.94 / 73.74 80.12 / 75.67 75.41 / 74.01 79.74 / 76.41 83.35 / 80.03 84.02 / 80.54 - / - - / - - / - - / - - / - en → tr 78.17 / 77.51 84.19 / 82.96 85.59 / 84.22 78.07 / 76.71 81.47 / 79.69 86.39 / 84.41 85.02 / 81.94 86.94 / 84.56 83.86 / 81.07 85.91 / 83.18 86.74 / 84.24 77.26 / 74.84 78.15 / 76.32 - / - - / - - / - en → ur 64.24 / 67.83 79.70 / 78.53 82.22 / 80.09 74.11 / 75.81 77.90 / 77.80 80.10 / 75.64 78.50 / 74.19 79.42 / 75.87 77.71 / 76.17 83.87 / 80.87 84.20 / 81.24 - / - - / - - / - - / - - / - en → uz 21.43 / 27.54 61.10 / 77.93 66.21 / 80.67 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → vi 85.85 / 80.70 89.29 / 85.05 89.94 / 85.41 84.79 / 81.36 86.57 / 82.77 91.55 / 86.26 89.82 / 82.81 91.51 / 86.08 88.49 / 83.33 91.35 / 85.72 91.46 / 85.99 83.08 / 78.64 84.15 / 80.67 - / - - / - - / - en → yue 80.80 / 78.28 85.26 / 82.28 86.61 / 83.50 - / - - / - 86.13 / 82.75 83.32 / 62.40 85.41 / 82.50 82.21 / 78.16 84.25 / 81.95 82.87 / 81.37 - / - - / - - / - - / - - / - en → zhs 82.13 / 77.89 85.61 / 81.45 86.71 / 82.43 82.42 / 78.72 84.11 / 80.32 89.13 / 83.51 86.67 / 81.59 88.49 / 83.22 85.73 / 81.21 88.61 / 83.04 88.76 / 83.18 84.02 / 80.43 85.58 / 82.48 79.41 / 75.57 83.00 / 79.49 83.22 / 80.00 en → zht 81.45 / 77.32 85.50 / 81.85 86.61 / 83.36 - / - - / - 89.36 / 83.29 87.04 / 81.30 87.88 / 83.74 85.75 / 81.33 87.66 / 83.41 88.53 / 83.68 - / - - / - 78.12 / 76.17 81.75 / 79.29 82.10 / 80.18 Table 6: Reference-free evaluation results (XCOMET / COMETKiwi) of baseline models (Part I) on the WMT24++ benchmark. Direction Google Translate NLLB-54.5B Gemini 2.5 Pro Gemini 3 Pro GPT-5 MiLMMT-46-1B-v0.1 MiLMMT-46-1B-v1.0 MiLMMT-46-4B-v0.1 MiLMMT-46-4B-v1.0 MiLMMT-46-12B-v0.1 MiLMMT-46-12B-v1.0 en → ar 86.44 / 81.97 - / - 84.85 / 81.08 85.10 / 81.20 84.36 / 81.56 74.32 / 71.17 78.72 / 75.15 82.17 / 78.59 86.67 / 81.80 83.68 / 80.37 89.44 / 84.33 en → az 75.04 / 78.02 - / - 78.83 / 81.72 80.01 / 82.17 80.36 / 82.29 66.15 / 69.98 73.65 / 75.98 77.81 / 79.42 83.99 / 83.98 80.36 / 81.29 87.76 / 86.56 en → bg 85.07 / 80.04 - / - 86.98 / 81.51 87.13 / 81.84 87.24 / 82.49 78.31 / 72.59 83.44 / 77.26 85.18 / 79.44 89.97 / 84.08 87.18 / 81.96 91.83 / 86.83 en → bn 76.65 / 76.64 - / - 81.98 / 79.63 84.39 / 81.09 82.14 / 80.01 73.35 / 72.87 80.44 / 77.38 81.13 / 79.21 87.39 / 81.83 82.97 / 80.67 89.00 / 83.30 en → ca 82.82 / 76.02 - / - 84.16 / 77.08 84.97 / 77.73 85.39 / 78.61 79.25 / 70.51 82.62 / 74.97 84.88 / 77.31 88.83 / 81.39 85.82 / 78.49 90.60 / 83.50 en → cs 83.42 / 77.94 - / - 85.82 / 80.02 85.95 / 80.36 86.08 / 81.57 74.07 / 66.96 79.72 / 73.77 82.84 / 77.04 87.89 / 82.89 85.99 / 80.36 90.08 / 85.44 en → da 90.57 / 81.72 - / - 91.88 / 83.64 92.15 / 84.16 92.06 / 84.59 86.69 / 76.36 89.50 / 79.45 90.86 / 81.85 93.55 / 85.42 92.14 / 83.36 95.05 / 87.05 en → de 94.57 / 81.13 - / - 93.93 / 80.31 93.84 / 80.70 93.87 / 81.08 89.73 / 73.16 91.87 / 75.70 93.32 / 79.43 94.81 / 81.13 94.11 / 80.88 95.83 / 82.72 en → el 83.85 / 78.92 - / - 86.04 / 81.08 85.76 / 80.96 86.33 / 81.45 75.94 / 70.33 79.82 / 74.29 84.67 / 78.69 88.74 / 82.62 85.97 / 81.03 91.04 / 85.24 en → es 89.20 / 79.21 - / - 89.32 / 79.50 89.46 / 79.65 89.86 / 80.50 85.69 / 74.59 88.29 / 78.24 89.02 / 79.17 92.18 / 82.46 89.79 / 80.35 93.52 / 84.39 en → fa 82.95 / 82.21 - / - 84.28 / 82.93 84.95 / 83.20 84.81 / 83.55 70.87 / 69.98 77.74 / 76.46 80.84 / 80.03 86.74 / 83.31 83.88 / 82.99 89.83 / 86.07 en → fi 88.18 / 85.06 - / - 90.53 / 87.55 90.76 / 87.38 90.23 / 88.09 75.47 / 72.56 79.05 / 75.77 86.58 / 83.34 90.13 / 86.72 89.23 / 85.77 92.83 / 89.55 en → fr 88.37 / 80.72 - / - 85.92 / 78.89 85.78 / 78.70 86.18 / 79.68 80.72 / 73.15 83.88 / 76.74 85.41 / 78.14 89.12 / 81.89 86.72 / 79.98 90.87 / 83.71 en → he 80.72 / 76.43 - / - 82.68 / 78.61 83.26 / 79.20 83.42 / 79.48 72.15 / 66.61 75.51 / 70.09 81.07 / 76.50 86.10 / 80.69 83.11 / 78.75 88.98 / 83.05 en → hi 84.30 / 74.35 - / - 79.63 / 72.28 81.11 / 73.33 81.02 / 72.74 68.30 / 66.08 75.43 / 67.59 76.95 / 71.28 84.66 / 72.50 78.71 / 72.65 87.43 / 74.67 en → hr 85.80 / 82.13 - / - 87.48 / 82.91 87.74 / 83.33 87.97 / 83.58 78.57 / 73.02 83.74 / 77.81 86.04 / 81.20 90.68 / 85.53 87.88 / 82.84 92.33 / 87.30 en → hu 86.37 / 83.60 - / - 88.66 / 85.48 88.95 / 85.20 89.45 / 86.08 75.42 / 71.75 79.77 / 75.75 85.50 / 82.16 90.04 / 85.74 88.13 / 84.92 92.35 / 88.20 en → id 90.46 / 84.30 - / - 89.55 / 83.29 89.59 / 83.12 89.86 / 83.29 84.81 / 77.44 89.02 / 81.12 88.70 / 81.60 92.67 / 85.58 89.75 / 82.48 94.04 / 86.73 en → it 87.68 / 80.30 - / - 86.98 / 79.38 87.53 / 79.56 87.94 / 80.65 83.21 / 73.62 86.62 / 78.19 88.13 / 79.58 91.32 / 83.55 88.69 / 80.86 92.37 / 85.20 en → ja 87.60 / 87.49 - / - 87.16 / 86.92 87.42 / 87.37 88.01 / 87.90 79.56 / 81.91 84.57 / 84.91 86.25 / 86.28 90.24 / 88.89 88.38 / 87.63 90.80 / 89.57 en → k 73.37 / 83.05 - / - 75.89 / 85.17 76.34 / 84.05 76.43 / 84.91 63.49 / 75.97 67.87 / 79.67 74.60 / 83.53 78.17 / 86.36 77.03 / 85.22 81.48 / 88.07 en → km 72.83 / 81.55 - / - 74.46 / 83.33 76.32 / 83.88 74.17 / 82.78 67.64 / 76.34 71.55 / 79.96 74.93 / 82.01 79.53 / 84.45 76.71 / 83.30 81.65 / 85.75 en → ko 90.76 / 86.91 - / - 88.69 / 85.01 89.34 / 85.89 88.85 / 86.06 80.71 / 79.32 85.25 / 82.66 87.27 / 84.30 90.46 / 86.54 89.07 / 86.01 92.43 / 88.02 en → lo 73.29 / 77.81 - / - 75.11 / 80.35 76.44 / 80.81 72.83 / 78.57 66.58 / 72.10 71.54 / 76.03 76.27 / 79.50 81.09 / 81.50 77.56 / 80.36 83.41 / 83.01 en → ms 82.57 / 76.26 - / - 86.49 / 80.14 86.24 / 79.79 86.75 / 80.42 83.15 / 75.98 87.19 / 79.63 86.21 / 79.31 91.89 / 84.13 87.10 / 80.22 93.51 / 85.64 en → my 70.34 / 80.67 - / - 72.74 / 83.01 73.61 / 83.07 72.24 / 82.92 59.62 / 73.65 66.31 / 77.34 72.25 / 82.11 78.40 / 84.23 75.30 / 83.95 80.98 / 85.58 en → nb 90.21 / 83.50 - / - 91.96 / 85.05 92.31 / 85.72 92.37 / 85.87 86.78 / 78.64 89.39 / 80.85 91.30 / 83.89 93.59 / 86.49 92.33 / 84.93 94.86 / 88.18 en → nl 93.72 / 85.21 - / - 92.18 / 83.21 92.14 / 83.43 92.42 / 84.25 87.32 / 76.36 89.77 / 79.31 91.46 / 81.90 93.62 / 83.98 92.75 / 83.27 94.82 / 85.92 en → pl 87.66 / 80.20 - / - 87.76 / 80.06 87.80 / 80.38 88.21 / 81.36 77.23 / 68.06 82.43 / 73.66 85.91 / 77.83 90.37 / 82.49 88.33 / 80.65 92.55 / 84.80 en → pt 91.25 / 82.85 - / - 89.28 / 80.68 89.21 / 80.75 89.84 / 81.82 86.88 / 77.10 89.58 / 80.34 89.53 / 81.25 92.72 / 84.68 90.89 / 82.48 93.76 / 86.38 en → ro 85.46 / 83.44 - / - 87.16 / 85.04 87.67 / 85.94 88.39 / 86.46 79.26 / 75.67 83.84 / 80.63 87.13 / 83.63 90.31 / 87.69 88.30 / 85.78 91.51 / 89.24 en → ru 87.83 / 82.89 - / - 85.95 / 80.63 86.21 / 80.57 86.27 / 81.29 77.44 / 70.78 82.96 / 76.78 84.16 / 78.82 88.88 / 82.75 85.86 / 81.44 90.49 / 85.04 en → sk 83.94 / 79.57 - / - 86.44 / 81.72 86.01 / 81.44 85.87 / 82.64 74.04 / 68.09 79.96 / 74.42 83.35 / 77.73 88.53 / 83.18 86.27 / 81.08 91.02 / 85.99 en → sl 83.83 / 79.54 - / - 85.59 / 81.15 85.84 / 80.96 86.45 / 81.72 74.38 / 68.05 80.02 / 74.57 83.35 / 77.91 88.14 / 82.69 85.93 / 80.35 90.89 / 85.47 en → sv 90.42 / 83.09 - / - 92.35 / 85.01 92.59 / 85.32 92.60 / 85.89 86.92 / 77.68 89.23 / 80.31 91.22 / 83.41 93.55 / 85.97 92.43 / 85.14 95.20 / 88.39 en → ta 68.86 / 75.71 - / - 75.64 / 80.09 78.68 / 81.25 74.21 / 79.33 60.90 / 68.78 68.15 / 73.61 72.24 / 77.97 78.42 / 80.04 74.90 / 80.21 82.41 / 82.30 en → th 90.25 / 84.97 - / - 87.17 / 82.56 88.45 / 83.07 87.65 / 83.38 77.58 / 75.15 83.03 / 79.43 85.41 / 81.49 90.07 / 84.52 86.98 / 83.22 92.02 / 86.25 en → tl 77.44 / 77.23 - / - 79.36 / 78.64 80.01 / 77.76 79.21 / 79.84 72.75 / 69.97 77.88 / 74.47 79.05 / 76.19 84.52 / 80.16 80.30 / 77.27 87.11 / 82.14 en → tr 85.63 / 83.93 - / - 82.57 / 81.15 82.34 / 80.98 82.53 / 81.49 72.29 / 71.77 79.07 / 77.64 80.38 / 79.14 86.40 / 84.19 83.15 / 81.36 89.32 / 86.01 en → ur 65.31 / 68.17 - / - 80.92 / 79.68 81.99 / 80.27 81.29 / 79.88 67.32 / 69.99 74.13 / 74.45 76.90 / 77.64 85.28 / 81.38 79.55 / 79.17 87.73 / 83.15 en → uz 64.85 / 78.22 - / - 67.89 / 80.47 68.92 / 80.94 68.93 / 80.71 58.77 / 73.11 63.98 / 77.57 68.52 / 79.45 73.68 / 83.11 71.25 / 81.16 77.41 / 84.92 en → vi 89.50 / 85.57 - / - 87.36 / 83.42 87.90 / 83.39 87.49 / 83.49 83.02 / 79.01 86.02 / 82.11 86.10 / 82.00 90.78 / 85.71 87.67 / 83.67 92.24 / 86.75 en → yue 66.14 / 67.40 - / - 73.98 / 74.38 73.80 / 74.14 77.25 / 77.78 72.65 / 71.61 74.69 / 74.80 76.29 / 76.17 82.70 / 80.77 78.04 / 77.78 83.80 / 82.53 en → zhs 86.36 / 82.02 - / - 84.35 / 80.39 85.15 / 80.70 85.02 / 80.72 79.02 / 75.24 82.76 / 78.93 83.07 / 79.14 87.92 / 82.62 84.59 / 81.00 89.49 / 84.01 en → zht 86.07 / 82.62 - / - 84.19 / 81.50 85.33 / 81.93 84.94 / 81.65 77.89 / 75.60 82.00 / 79.30 82.85 / 80.22 88.01 / 83.44 84.38 / 82.01 89.33 / 84.67 Table 7: Reference-free evaluation results (XCOMET / COMETKiwi) of baseline models (Part I) and MiLMMT models on the WMT24++ benchmark. Direction TranslateGemma-4B TranslateGemma-12B TranslateGemma-27B GemmaX2-28-2B GemmaX2-28-9B Hunyuan-MT-7B HY-MT1.5-1.8B HY-MT1.5-7B HY-MT2-1.8B HY-MT2-7B HY-MT2-30B-A3B Seed-X-Instruct-7B Seed-X-PPO-7B Tower-Plus-2B Tower-Plus-9B Tower-Plus-72B en → ar 93.54 / 87.09 95.66 / 89.84 96.38 / 90.84 91.90 / 85.63 94.20 / 87.97 96.24 / 90.07 93.88 / 86.84 96.05 / 90.22 93.32 / 86.88 95.60 / 89.57 95.70 / 89.69 93.34 / 87.04 93.78 / 87.76 - / - - / - - / - ar → en 87.46 / 84.55 90.50 / 85.89 91.93 / 86.59 91.31 / 86.44 92.65 / 87.19 90.88 / 85.58 88.41 / 83.29 91.20 / 85.75 89.88 / 85.05 92.30 / 86.81 92.62 / 86.69 90.08 / 85.88 92.70 / 86.99 - / - - / - - / - en → az 50.49 / 54.71 87.63 / 87.30 92.72 / 91.02 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - az → en 67.08 / 83.64 76.62 / 86.34 81.93 / 87.87 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → bg 94.35 / 89.46 97.23 / 93.44 97.88 / 94.36 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - bg → en 91.52 / 89.06 93.40 / 90.01 94.93 / 90.79 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → bn 90.38 / 81.97 93.58 / 85.14 94.19 / 85.99 91.40 / 84.35 93.15 / 86.05 94.81 / 86.03 92.21 / 81.61 95.05 / 86.60 91.99 / 82.94 94.80 / 86.89 95.37 / 87.59 - / - - / - - / - - / - - / - bn → en 83.27 / 82.49 87.26 / 84.17 89.70 / 85.15 88.99 / 85.76 90.71 / 86.67 83.57 / 81.19 85.63 / 80.88 85.57 / 82.73 85.82 / 83.39 88.70 / 85.33 90.36 / 86.00 - / - - / - - / - - / - - / - en → ca 93.41 / 86.49 96.70 / 91.16 96.85 / 91.74 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - ca → en 91.23 / 90.53 93.42 / 91.84 94.83 / 92.26 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → cs 92.31 / 88.02 95.96 / 92.97 96.63 / 94.28 94.08 / 88.52 95.67 / 91.28 96.30 / 92.40 92.77 / 87.03 96.11 / 92.23 92.99 / 86.87 96.25 / 91.89 96.66 / 92.79 95.12 / 91.05 96.01 / 91.83 93.43 / 87.92 95.68 / 91.76 95.10 / 90.73 cs → en 92.65 / 90.36 93.93 / 91.39 95.19 / 91.99 95.39 / 91.77 96.12 / 92.16 94.12 / 90.84 92.91 / 88.80 93.99 / 90.86 94.15 / 90.32 95.79 / 92.02 95.67 / 91.85 91.50 / 90.54 95.85 / 92.22 94.65 / 91.33 95.51 / 92.23 95.83 / 92.13 en → da 96.86 / 91.47 98.21 / 93.86 98.62 / 94.75 - / - - / - - / - - / - - / - - / - - / - - / - 97.71 / 92.56 97.75 / 92.86 96.87 / 90.90 97.76 / 92.06 97.18 / 91.61 da → en 94.88 / 92.30 95.57 / 92.79 96.61 / 93.29 - / - - / - - / - - / - - / - - / - - / - - / - 95.43 / 92.58 97.33 / 93.74 96.52 / 93.28 97.06 / 93.60 97.25 / 93.71 en → de 98.00 / 85.26 98.66 / 87.16 98.94 / 88.25 97.88 / 85.32 98.37 / 86.99 98.73 / 87.42 97.98 / 84.32 98.69 / 87.49 97.80 / 84.32 98.55 / 86.96 98.76 / 87.99 98.45 / 86.75 98.28 / 87.17 97.75 / 84.73 98.51 / 87.15 98.58 / 87.22 de → en 95.43 / 89.92 95.98 / 90.71 96.62 / 91.25 96.86 / 91.50 97.06 / 91.63 96.18 / 90.43 95.80 / 88.68 95.96 / 90.35 95.79 / 90.41 96.88 / 91.45 96.77 / 91.30 95.64 / 90.64 97.01 / 91.68 96.28 / 91.04 96.78 / 91.42 96.93 / 91.56 en → el 92.77 / 86.77 96.12 / 90.66 96.90 / 92.41 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - el → en 89.96 / 88.16 92.20 / 89.38 93.60 / 90.10 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → es 97.52 / 91.08 97.79 / 92.01 98.11 / 92.32 97.05 / 89.59 97.57 / 90.63 98.07 / 91.88 97.76 / 90.38 97.98 / 91.90 97.29 / 89.69 97.65 / 91.16 98.01 / 91.77 97.15 / 89.41 97.32 / 89.73 96.89 / 89.03 97.83 / 91.07 97.73 / 91.07 es → en 94.11 / 90.21 95.51 / 90.83 96.22 / 91.32 96.03 / 91.20 96.41 / 91.42 95.42 / 90.22 94.92 / 88.25 95.53 / 90.40 95.43 / 90.46 96.06 / 91.14 96.33 / 91.28 91.35 / 88.18 96.61 / 91.48 95.89 / 90.85 96.08 / 91.36 96.17 / 91.39 en → fa 91.29 / 87.76 95.76 / 91.96 96.65 / 92.71 92.03 / 88.45 94.86 / 91.37 94.11 / 89.93 91.53 / 86.43 94.40 / 90.03 91.44 / 86.99 95.36 / 91.49 95.85 / 92.22 - / - - / - - / - - / - - / - fa → en 87.32 / 84.38 90.78 / 85.68 93.11 / 87.02 92.94 / 87.32 93.73 / 87.82 90.49 / 85.05 87.55 / 82.54 90.97 / 85.42 88.93 / 84.65 92.47 / 87.07 93.13 / 87.37 - / - - / - - / - - / - - / - en → fi 92.11 / 90.21 95.94 / 94.89 97.17 / 96.23 - / - - / - - / - - / - - / - - / - - / - - / - 95.31 / 93.49 96.20 / 94.52 92.76 / 90.22 96.39 / 94.64 94.84 / 92.68 fi → en 89.48 / 89.20 92.60 / 90.42 94.23 / 90.84 - / - - / - - / - - / - - / - - / - - / - - / - 92.15 / 90.00 94.72 / 90.81 92.35 / 89.89 94.43 / 90.99 94.60 / 90.97 en → fr 95.72 / 91.57 96.89 / 92.93 96.97 / 93.05 95.70 / 90.38 96.94 / 91.81 96.80 / 92.47 95.45 / 89.92 96.97 / 92.68 95.33 / 89.48 96.98 / 91.92 97.10 / 92.25 96.62 / 91.11 96.77 / 91.17 95.25 / 89.22 96.76 / 91.71 97.02 / 92.02 fr → en 93.51 / 91.37 94.46 / 92.00 95.20 / 92.39 96.10 / 92.46 96.20 / 92.73 95.03 / 91.50 94.08 / 89.63 94.91 / 91.56 95.35 / 91.56 96.05 / 92.45 95.85 / 92.25 92.46 / 90.56 96.21 / 92.70 95.78 / 92.29 95.90 / 92.54 95.89 / 92.64 en → he 87.79 / 79.92 94.14 / 87.65 95.68 / 89.66 91.47 / 84.40 94.21 / 87.82 90.18 / 81.22 86.99 / 77.13 90.45 / 81.86 88.62 / 79.61 94.10 / 87.21 95.13 / 88.30 - / - - / - - / - - / - - / - he → en 89.92 / 85.71 93.24 / 87.51 94.62 / 88.50 94.37 / 88.29 95.24 / 89.01 91.09 / 85.77 89.52 / 83.63 91.59 / 86.24 89.79 / 85.52 94.61 / 88.29 94.90 / 88.62 - / - - / - - / - - / - - / - en → hi 89.14 / 75.25 92.66 / 78.04 93.71 / 79.11 86.70 / 76.38 89.61 / 78.38 94.14 / 76.46 91.81 / 72.28 94.27 / 77.82 89.95 / 75.15 93.96 / 79.22 94.32 / 79.70 - / - - / - 83.44 / 74.54 86.97 / 77.13 86.65 / 76.52 hi → en 90.52 / 77.12 93.06 / 78.38 94.17 / 80.09 93.71 / 81.86 94.89 / 82.57 92.25 / 77.84 91.51 / 76.44 93.05 / 78.55 91.70 / 79.79 94.13 / 81.61 95.02 / 81.51 - / - - / - 92.56 / 80.92 94.73 / 82.03 94.60 / 81.87 en → hr 92.24 / 85.92 97.12 / 93.10 98.02 / 94.65 - / - - / - - / - - / - - / - - / - - / - - / - 94.28 / 88.13 95.73 / 90.77 - / - - / - - / - hr → en 91.27 / 89.68 93.89 / 91.28 95.14 / 91.72 - / - - / - - / - - / - - / - - / - - / - - / - 88.15 / 88.54 95.06 / 91.65 - / - - / - - / - en → hu 87.86 / 82.25 95.14 / 91.78 97.21 / 94.72 - / - - / - - / - - / - - / - - / - - / - - / - 96.15 / 92.40 96.97 / 93.96 93.58 / 89.14 96.43 / 93.10 95.44 / 91.55 hu → en 89.51 / 89.57 93.05 / 91.48 94.59 / 92.17 - / - - / - - / - - / - - / - - / - - / - - / - 91.75 / 90.96 95.20 / 92.45 93.75 / 91.61 94.64 / 92.67 95.49 / 92.61 en → id 96.98 / 91.88 97.82 / 93.27 98.17 / 93.72 96.27 / 91.01 96.75 / 91.40 98.20 / 93.18 97.00 / 91.07 98.15 / 93.25 96.58 / 91.22 97.76 / 92.83 97.84 / 93.21 95.88 / 90.04 95.98 / 90.50 - / - - / - - / - id → en 95.13 / 88.14 95.55 / 88.48 96.41 / 89.15 96.44 / 89.48 96.65 / 89.76 95.59 / 87.88 94.44 / 85.80 95.51 / 87.90 95.26 / 88.10 96.46 / 89.21 96.41 / 89.37 94.95 / 88.56 96.69 / 89.46 - / - - / - - / - en → it 96.92 / 91.66 97.54 / 93.04 97.77 / 93.78 96.65 / 90.70 97.10 / 91.61 98.12 / 93.15 96.56 / 90.46 97.90 / 93.08 96.47 / 89.81 97.46 / 92.15 97.87 / 92.91 96.73 / 90.99 96.62 / 90.75 96.12 / 89.92 97.44 / 92.06 97.31 / 91.94 it → en 94.72 / 91.69 94.95 / 92.06 95.68 / 92.54 96.23 / 92.74 96.73 / 93.02 94.90 / 91.57 94.43 / 89.65 95.02 / 91.80 95.31 / 91.77 96.03 / 92.69 96.08 / 92.72 91.30 / 90.04 96.62 / 93.02 95.65 / 92.55 96.02 / 93.02 96.22 / 93.00 en → ja 91.36 / 93.58 93.30 / 95.21 94.05 / 95.40 93.50 / 94.01 94.82 / 95.12 95.71 / 95.74 94.56 / 94.37 95.84 / 95.69 94.53 / 94.20 95.63 / 95.60 95.59 / 95.77 92.56 / 93.23 94.54 / 94.68 92.54 / 93.25 94.62 / 95.11 94.99 / 95.07 ja → en 91.24 / 87.18 93.90 / 88.32 95.47 / 89.22 94.64 / 88.93 95.76 / 89.56 94.98 / 88.64 93.62 / 86.93 95.07 / 88.71 94.30 / 88.10 95.58 / 89.28 95.77 / 89.36 91.64 / 87.27 95.44 / 89.35 93.52 / 88.15 95.02 / 89.25 95.29 / 89.36 en → k 52.27 / 67.28 83.71 / 90.74 90.39 / 93.87 - / - - / - 45.97 / 76.49 66.47 / 65.09 45.84 / 77.00 46.80 / 61.73 57.17 / 75.23 48.06 / 77.72 - / - - / - - / - - / - - / - k → en 46.93 / 82.77 57.58 / 85.89 63.37 / 87.04 - / - - / - 31.73 / 72.05 22.09 / 64.42 34.93 / 74.27 23.17 / 64.93 50.29 / 80.40 66.01 / 86.84 - / - - / - - / - - / - - / - en → km 39.67 / 50.32 76.04 / 83.31 82.54 / 86.94 83.80 / 87.40 86.43 / 89.11 81.22 / 85.92 81.58 / 85.86 80.30 / 85.99 85.37 / 88.43 88.20 / 90.15 87.95 / 90.57 - / - - / - - / - - / - - / - km → en 49.77 / 83.57 59.12 / 86.19 64.48 / 87.19 68.17 / 87.04 70.90 / 87.81 59.03 / 84.34 64.30 / 84.58 59.86 / 85.37 66.72 / 86.34 68.89 / 87.43 67.53 / 87.68 - / - - / - - / - - / - - / - en → ko 94.06 / 90.06 96.72 / 92.59 96.93 / 92.93 94.13 / 90.22 95.86 / 92.03 97.49 / 93.11 96.05 / 91.39 97.21 / 93.01 95.37 / 91.37 96.98 / 93.04 96.94 / 93.06 91.23 / 87.78 93.49 / 89.76 93.30 / 89.40 96.11 / 91.94 95.71 / 91.56 ko → en 90.33 / 86.34 93.14 / 87.48 94.47 / 88.54 93.74 / 88.41 95.03 / 89.12 93.83 / 87.73 92.82 / 86.42 94.15 / 87.92 93.27 / 87.90 95.09 / 88.80 95.12 / 88.76 91.28 / 87.04 94.10 / 88.43 92.11 / 87.85 94.35 / 88.79 94.76 / 89.04 en → lo 41.02 / 51.67 79.54 / 81.77 86.80 / 86.01 84.54 / 84.54 88.40 / 87.15 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - lo → en 62.71 / 79.36 74.09 / 82.26 77.71 / 83.28 79.43 / 83.32 82.07 / 84.10 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → ms 93.94 / 87.28 96.60 / 90.97 96.80 / 90.87 95.16 / 88.76 95.68 / 89.39 97.04 / 90.80 96.68 / 89.75 97.29 / 91.35 95.92 / 89.53 96.84 / 90.88 97.21 / 91.43 93.74 / 86.70 94.71 / 88.28 - / - - / - - / - ms → en 93.43 / 85.89 94.92 / 86.92 95.98 / 87.72 96.15 / 88.24 96.31 / 88.37 94.86 / 86.57 93.94 / 84.49 95.38 / 86.76 94.52 / 86.37 96.16 / 88.06 96.42 / 88.07 94.02 / 86.75 96.31 / 88.11 - / - - / - - / - en → my 44.84 / 61.03 69.48 / 81.47 79.44 / 86.61 79.54 / 86.53 86.80 / 89.94 78.03 / 86.18 79.54 / 84.88 78.94 / 86.63 83.38 / 88.08 88.72 / 90.95 88.13 / 90.78 - / - - / - - / - - / - - / - my → en 45.66 / 81.57 58.84 / 85.79 65.43 / 87.35 66.26 / 87.26 70.33 / 88.62 57.03 / 82.99 63.99 / 84.48 60.16 / 84.84 65.97 / 86.69 69.91 / 88.29 68.31 / 88.54 - / - - / - - / - - / - - / - en → nb 96.92 / 92.59 98.30 / 94.96 98.58 / 95.30 - / - - / - - / - - / - - / - - / - - / - - / - 95.55 / 89.99 96.70 / 91.74 95.95 / 90.71 95.76 / 90.47 95.33 / 89.99 nb → en 94.33 / 90.63 94.96 / 91.21 96.01 / 92.10 - / - - / - - / - - / - - / - - / - - / - - / - 94.14 / 91.28 96.11 / 92.25 95.59 / 91.75 96.55 / 92.39 96.41 / 92.49 en → nl 97.15 / 90.05 97.99 / 92.52 98.53 / 93.45 97.26 / 90.53 98.03 / 92.21 97.40 / 90.10 96.36 / 87.41 97.40 / 90.14 96.32 / 88.30 97.97 / 91.64 98.31 / 92.59 97.84 / 90.83 98.00 / 92.23 96.93 / 90.04 98.12 / 92.52 98.15 / 92.27 nl → en 94.81 / 91.19 95.49 / 91.91 96.53 / 92.49 96.69 / 92.32 97.08 / 92.79 95.14 / 90.54 94.58 / 88.93 95.26 / 90.95 95.39 / 90.89 96.85 / 92.46 96.79 / 92.39 90.58 / 88.81 97.12 / 92.66 96.31 / 92.13 96.75 / 92.67 96.85 / 92.81 en → pl 95.00 / 88.15 97.23 / 92.02 97.88 / 93.12 94.87 / 87.30 96.88 / 90.39 94.78 / 87.29 92.97 / 83.67 95.28 / 87.79 93.46 / 83.78 96.90 / 90.45 97.76 / 91.82 96.30 / 89.27 96.98 / 90.85 94.94 / 87.11 97.20 / 91.41 96.89 / 90.35 pl → en 93.05 / 77.62 93.75 / 78.30 95.25 / 78.71 95.47 / 78.62 96.28 / 78.87 93.09 / 77.19 92.51 / 76.13 93.47 / 77.50 93.21 / 77.15 95.39 / 78.53 95.34 / 78.73 87.36 / 76.09 95.59 / 78.80 94.52 / 78.27 95.66 / 78.94 95.64 / 78.92 en → pt 97.58 / 92.27 98.10 / 93.27 98.26 / 93.55 97.33 / 91.42 97.74 / 92.05 98.16 / 92.82 97.32 / 91.24 98.10 / 92.93 97.18 / 91.05 97.78 / 92.20 97.94 / 92.63 97.18 / 90.76 97.13 / 90.99 96.64 / 90.48 97.48 / 91.97 97.68 / 92.31 pt → en 94.77 / 92.42 95.39 / 92.95 96.01 / 93.48 97.00 / 93.91 97.13 / 94.18 95.33 / 92.15 94.98 / 90.74 95.62 / 92.41 95.59 / 92.83 96.60 / 93.72 96.57 / 93.70 95.37 / 93.10 96.88 / 93.88 96.25 / 93.52 96.89 / 93.98 96.89 / 93.94 en → ro 94.60 / 94.14 96.90 / 96.32 97.64 / 96.97 - / - - / - - / - - / - - / - - / - - / - - / - 95.92 / 94.89 96.42 / 95.22 94.77 / 92.92 97.06 / 95.53 96.37 / 94.82 ro → en 91.85 / 96.34 92.75 / 96.76 94.58 / 97.21 - / - - / - - / - - / - - / - - / - - / - - / - 87.50 / 95.56 95.41 / 97.35 94.09 / 96.92 95.13 / 97.35 95.47 / 97.34 en → ru 95.22 / 89.94 96.82 / 92.38 97.26 / 93.05 93.40 / 87.06 95.75 / 90.42 96.54 / 91.53 94.37 / 87.80 96.46 / 91.49 93.25 / 86.44 96.21 / 90.99 96.81 / 92.17 96.06 / 90.92 96.21 / 91.10 93.79 / 87.26 96.09 / 91.18 96.33 / 91.48 ru → en 92.86 / 85.52 93.87 / 86.20 95.31 / 87.00 94.96 / 87.00 95.66 / 87.17 94.52 / 86.16 93.48 / 84.24 94.80 / 86.31 94.00 / 85.68 95.24 / 86.87 95.47 / 86.93 92.26 / 85.34 95.76 / 87.15 94.19 / 86.36 95.31 / 86.98 95.37 / 87.29 en → sk 89.69 / 84.33 95.78 / 92.54 96.47 / 94.08 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - sk → en 90.86 / 89.77 92.92 / 91.11 94.45 / 91.82 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → sl 85.69 / 76.95 94.58 / 89.07 96.50 / 92.45 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - sl → en 88.76 / 88.87 92.35 / 90.52 94.06 / 91.40 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → sv 96.98 / 91.85 98.45 / 94.50 98.75 / 95.26 - / - - / - - / - - / - - / - - / - - / - - / - 97.71 / 92.79 97.77 / 93.25 96.81 / 91.43 97.88 / 93.59 97.66 / 92.80 sv → en 94.72 / 91.90 95.34 / 92.36 96.06 / 92.86 - / - - / - - / - - / - - / - - / - - / - - / - 92.86 / 91.00 96.53 / 93.16 95.87 / 92.63 96.54 / 93.23 96.74 / 93.30 en → ta 84.63 / 81.77 89.01 / 85.21 89.78 / 86.01 - / - - / - 82.73 / 79.84 84.28 / 78.38 84.10 / 80.65 83.45 / 81.18 89.26 / 86.21 89.98 / 86.54 - / - - / - - / - - / - - / - ta → en 76.92 / 72.73 82.10 / 74.41 86.09 / 75.55 - / - - / - 76.18 / 72.26 76.86 / 71.75 79.04 / 73.16 79.31 / 73.85 86.85 / 76.16 87.00 / 76.46 - / - - / - - / - - / - - / - en → th 93.28 / 87.47 96.56 / 90.70 96.92 / 91.26 93.63 / 88.23 95.87 / 90.01 96.87 / 91.02 94.25 / 88.06 96.67 / 91.06 94.02 / 88.18 96.59 / 91.00 96.80 / 91.10 92.95 / 87.18 93.86 / 88.13 - / - - / - - / - th → en 91.44 / 85.42 93.72 / 86.51 95.13 / 87.44 94.23 / 87.41 95.48 / 88.22 93.93 / 86.16 92.72 / 84.10 94.36 / 86.51 92.82 / 85.72 95.28 / 87.88 95.27 / 87.95 89.89 / 85.20 94.83 / 87.67 - / - - / - - / - en → tl 89.88 / 85.89 92.70 / 88.89 92.45 / 88.89 89.82 / 86.25 91.53 / 87.45 87.53 / 83.83 90.38 / 85.46 87.11 / 84.26 90.48 / 86.02 92.78 / 88.83 93.10 / 89.18 - / - - / - - / - - / - - / - tl → en 89.18 / 87.21 91.38 / 88.23 92.92 / 89.04 92.28 / 89.10 93.10 / 89.53 90.87 / 87.52 90.21 / 85.82 91.08 / 87.82 90.61 / 87.88 92.32 / 89.09 92.95 / 89.08 - / - - / - - / - - / - - / - en → tr 91.38 / 88.33 94.72 / 92.43 95.75 / 93.48 91.96 / 88.80 94.22 / 91.22 95.46 / 92.74 94.11 / 90.89 95.83 / 93.10 93.11 / 89.86 95.31 / 92.40 96.07 / 93.09 91.85 / 88.27 92.89 / 89.47 - / - - / - - / - tr → en 85.81 / 87.56 88.60 / 89.07 91.10 / 90.04 91.49 / 90.06 92.50 / 90.83 88.87 / 88.18 88.13 / 86.91 89.10 / 88.53 90.17 / 88.89 92.09 / 90.40 92.24 / 90.42 87.61 / 88.70 92.03 / 90.23 - / - - / - - / - en → ur 77.97 / 73.20 90.72 / 83.77 94.15 / 86.31 88.21 / 82.75 91.58 / 85.30 89.87 / 81.64 88.46 / 79.21 90.24 / 82.18 88.61 / 81.10 93.46 / 86.17 93.93 / 86.71 - / - - / - - / - - / - - / - ur → en 80.27 / 80.75 86.01 / 83.37 89.42 / 85.04 89.04 / 85.81 90.35 / 86.75 85.36 / 82.14 81.71 / 79.85 86.01 / 83.00 81.51 / 82.31 89.33 / 85.76 89.71 / 86.01 - / - - / - - / - - / - - / - en → uz 22.48 / 35.37 81.39 / 88.51 86.84 / 91.38 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - uz → en 48.99 / 81.22 58.78 / 84.87 64.69 / 86.15 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → vi 94.87 / 91.46 96.35 / 93.19 97.03 / 93.63 94.43 / 91.39 95.43 / 92.45 97.10 / 93.38 95.85 / 91.02 96.98 / 93.50 94.72 / 91.27 96.68 / 93.30 96.82 / 93.61 93.57 / 90.14 94.24 / 90.95 - / - - / - - / - vi → en 93.60 / 87.97 94.86 / 88.99 95.77 / 89.66 95.42 / 89.74 96.14 / 90.20 95.19 / 88.67 94.41 / 86.71 95.24 / 88.75 94.42 / 88.51 95.94 / 89.94 95.96 / 90.03 92.21 / 87.21 96.00 / 89.82 - / - - / - - / - en → yue 90.43 / 87.06 93.92 / 89.65 94.24 / 90.45 - / - - / - 92.50 / 89.29 89.13 / 70.57 92.04 / 89.16 90.04 / 86.18 92.74 / 89.45 92.15 / 89.18 - / - - / - - / - - / - - / - yue → en 93.92 / 84.78 95.61 / 86.09 95.94 / 86.66 - / - - / - 95.85 / 86.77 94.27 / 84.67 95.99 / 86.80 94.50 / 85.66 95.98 / 86.98 95.98 / 86.98 - / - - / - - / - - / - - / - en → zhs 91.98 / 87.14 94.05 / 88.92 94.45 / 89.86 92.50 / 87.66 93.21 / 88.48 95.56 / 89.93 93.68 / 88.42 95.04 / 89.84 93.08 / 88.03 95.06 / 89.91 95.11 / 90.01 93.41 / 88.63 94.31 / 89.93 90.60 / 85.96 92.33 / 88.05 93.36 / 88.95 zhs → en 96.37 / 86.48 97.02 / 87.03 97.34 / 87.76 97.01 / 87.46 97.18 / 87.85 97.26 / 87.51 96.43 / 85.71 97.15 / 87.42 96.73 / 87.09 97.46 / 87.64 97.62 / 87.92 96.63 / 87.03 97.07 / 87.65 96.54 / 87.16 97.08 / 87.69 97.28 / 87.79 en → zht 91.44 / 87.00 93.92 / 89.13 94.20 / 90.50 - / - - / - 95.50 / 89.93 94.08 / 88.35 94.45 / 90.40 93.29 / 88.17 94.70 / 90.58 94.86 / 90.72 - / - - / - 89.54 / 86.35 92.11 / 88.56 92.80 / 89.25 zht → en 94.74 / 86.14 96.03 / 87.05 96.29 / 87.53 - / - - / - 96.10 / 87.43 95.09 / 85.77 96.08 / 87.28 95.16 / 86.96 96.06 / 87.55 96.30 / 87.58 - / - - / - 95.17 / 86.84 96.27 / 87.68 96.35 / 87.98 Table 8: Reference-free English-centric evaluation results (XCOMET / COMETKiwi) of baseline models (Part I) on the FLORES+ benchmark. Direction Google Translate NLLB-54.5B Gemini 2.5 Pro Gemini 3 Pro GPT-5 MiLMMT-46-1B-v0.1 MiLMMT-46-1B-v1.0 MiLMMT-46-4B-v0.1 MiLMMT-46-4B-v1.0 MiLMMT-46-12B-v0.1 MiLMMT-46-12B-v1.0 en → ar 95.05 / 89.48 91.61 / 84.52 95.64 / 89.85 95.68 / 90.03 94.83 / 89.87 89.03 / 82.35 91.63 / 85.10 93.67 / 87.88 95.49 / 89.82 94.77 / 88.90 96.96 / 91.53 ar → en 92.15 / 87.00 90.63 / 85.83 92.75 / 87.02 92.56 / 87.07 92.00 / 86.93 90.24 / 85.76 91.34 / 86.04 92.39 / 86.93 92.66 / 87.06 93.11 / 87.28 93.35 / 87.37 en → az 88.43 / 86.98 85.93 / 82.51 93.82 / 92.04 94.15 / 92.24 94.28 / 92.14 84.73 / 83.88 88.48 / 87.01 92.15 / 90.09 94.86 / 92.54 93.63 / 91.42 96.67 / 93.96 az → en 85.50 / 88.87 81.21 / 87.65 86.56 / 88.83 86.65 / 88.77 86.08 / 88.74 80.83 / 86.51 83.97 / 86.94 86.01 / 88.53 87.50 / 88.38 86.60 / 88.84 88.19 / 88.80 en → bg 95.99 / 91.09 93.16 / 86.35 96.79 / 93.02 97.26 / 93.28 96.10 / 93.18 92.30 / 86.17 94.08 / 88.79 96.02 / 91.61 97.41 / 94.00 96.73 / 92.41 98.28 / 95.08 bg → en 95.26 / 91.09 92.30 / 89.50 95.36 / 91.09 95.47 / 91.16 94.87 / 91.10 94.17 / 90.26 94.80 / 90.39 95.64 / 91.17 96.00 / 91.33 95.70 / 91.37 95.89 / 91.39 en → bn 90.40 / 84.17 87.81 / 80.98 92.82 / 85.51 94.31 / 86.87 91.56 / 85.61 86.95 / 81.05 91.83 / 83.94 92.52 / 85.61 95.51 / 87.45 93.64 / 86.36 96.69 / 88.49 bn → en 90.65 / 86.72 87.64 / 85.56 90.92 / 86.64 91.11 / 86.85 90.77 / 86.65 86.46 / 84.91 88.56 / 85.21 89.81 / 86.39 91.06 / 86.30 90.99 / 86.81 91.94 / 86.78 en → ca 95.14 / 88.88 93.18 / 84.47 95.67 / 89.74 95.47 / 89.40 95.31 / 90.48 94.28 / 86.68 95.81 / 89.22 95.70 / 89.39 97.34 / 92.20 96.18 / 90.03 97.96 / 93.42 ca → en 95.20 / 92.37 92.78 / 90.99 95.10 / 92.60 95.12 / 92.59 94.63 / 92.60 94.75 / 91.99 94.57 / 91.94 95.18 / 92.65 95.42 / 92.74 95.66 / 92.94 95.74 / 92.91 en → cs 95.45 / 90.95 91.45 / 86.98 96.22 / 92.36 96.54 / 92.77 96.48 / 93.05 90.61 / 83.96 93.09 / 88.28 95.59 / 91.06 96.78 / 93.37 96.32 / 92.32 97.50 / 95.13 cs → en 95.59 / 92.14 93.20 / 90.60 95.74 / 91.89 95.74 / 91.94 95.17 / 92.16 94.28 / 90.88 95.05 / 91.19 95.76 / 91.98 96.20 / 92.24 96.04 / 92.32 96.41 / 92.49 en → da 97.66 / 92.49 96.26 / 89.10 98.34 / 93.57 98.37 / 93.80 98.10 / 94.01 96.12 / 89.10 97.01 / 90.72 97.77 / 92.65 98.58 / 94.40 98.22 / 93.60 98.96 / 95.25 da → en 97.01 / 93.73 94.70 / 91.94 97.03 / 93.56 97.23 / 93.65 96.67 / 93.52 96.48 / 93.08 96.75 / 93.30 97.36 / 93.79 97.63 / 93.82 97.36 / 93.81 97.56 / 93.89 en → de 98.77 / 87.88 97.64 / 83.49 98.49 / 87.48 98.66 / 87.91 98.01 / 87.27 96.93 / 81.77 97.57 / 83.60 98.41 / 86.63 98.75 / 87.57 98.58 / 87.35 99.10 / 88.99 de → en 96.79 / 91.41 95.46 / 90.17 96.68 / 91.25 96.79 / 91.27 96.38 / 91.28 96.04 / 90.58 96.29 / 90.69 96.71 / 91.36 97.17 / 91.69 96.96 / 91.55 97.34 / 91.76 en → el 95.24 / 89.95 93.03 / 84.24 96.28 / 91.31 96.19 / 91.44 95.57 / 91.44 90.34 / 84.29 92.22 / 86.04 95.14 / 90.09 96.72 / 91.98 95.85 / 91.24 97.59 / 93.15 el → en 94.42 / 90.65 93.10 / 89.57 94.26 / 90.54 94.41 / 90.54 93.89 / 90.53 92.61 / 89.44 93.69 / 89.70 94.60 / 90.62 95.08 / 90.68 94.77 / 90.81 95.10 / 90.95 en → es 97.30 / 90.46 95.88 / 86.63 97.50 / 90.48 97.34 / 90.38 97.27 / 91.13 96.42 / 88.53 97.43 / 90.27 97.46 / 90.42 98.42 / 92.56 97.81 / 91.13 98.81 / 93.48 es → en 95.88 / 91.16 92.47 / 87.66 95.91 / 91.17 96.10 / 91.10 95.67 / 91.20 95.67 / 90.75 96.10 / 91.04 96.38 / 91.36 96.59 / 91.45 96.56 / 91.50 96.84 / 91.72 en → fa 95.04 / 91.46 90.52 / 86.06 95.69 / 92.10 95.58 / 92.10 94.41 / 92.10 85.79 / 82.45 89.65 / 85.86 93.67 / 89.97 95.35 / 91.47 94.92 / 91.51 96.92 / 93.32 fa → en 93.83 / 87.95 90.75 / 86.40 94.07 / 87.76 93.84 / 87.73 94.18 / 87.79 90.17 / 85.89 91.84 / 86.51 93.27 / 87.46 93.91 / 87.66 94.09 / 87.99 94.57 / 88.11 en → fi 96.28 / 94.71 92.46 / 90.76 97.35 / 96.02 97.46 / 96.28 96.89 / 96.40 87.34 / 84.07 89.71 / 86.44 95.38 / 93.47 96.83 / 95.40 96.62 / 95.04 97.96 / 96.74 fi → en 94.67 / 91.00 91.37 / 89.66 94.88 / 90.81 94.82 / 90.85 94.25 / 90.76 92.16 / 89.47 93.13 / 89.59 94.96 / 90.92 95.27 / 91.02 95.54 / 91.18 95.86 / 91.16 en → fr 96.50 / 91.46 93.79 / 86.71 96.43 / 91.42 96.63 / 91.64 96.34 / 91.79 94.53 / 88.13 95.19 / 89.85 96.49 / 91.13 97.33 / 93.22 97.03 / 92.00 98.06 / 94.11 fr → en 95.46 / 92.25 94.34 / 91.25 95.46 / 92.23 95.73 / 92.05 95.31 / 92.32 95.80 / 92.24 96.16 / 92.52 96.43 / 92.74 96.59 / 92.90 96.31 / 92.72 96.60 / 92.93 en → he 93.68 / 86.87 91.81 / 83.62 94.60 / 88.54 94.90 / 88.83 93.51 / 89.08 86.04 / 77.77 87.35 / 78.58 93.32 / 86.35 95.09 / 88.32 94.69 / 88.43 96.70 / 90.58 he → en 95.08 / 88.98 91.81 / 86.96 95.22 / 88.83 95.35 / 88.89 94.21 / 88.77 91.86 / 87.09 92.52 / 87.28 95.17 / 88.76 95.09 / 88.86 95.58 / 89.11 95.53 / 89.22 en → hi 92.01 / 79.33 83.66 / 75.63 91.30 / 78.96 92.64 / 79.50 91.93 / 78.18 82.47 / 73.39 88.77 / 75.10 89.00 / 77.77 93.61 / 78.40 91.04 / 79.36 95.25 / 80.00 hi → en 94.53 / 81.74 94.04 / 81.55 95.14 / 82.28 95.14 / 82.72 94.88 / 82.06 92.11 / 81.16 92.60 / 81.31 94.61 / 82.38 95.16 / 82.28 95.01 / 82.61 95.41 / 82.24 en → hr 97.24 / 93.12 94.04 / 87.40 97.83 / 93.90 97.88 / 93.96 96.63 / 94.21 93.07 / 86.46 94.71 / 89.17 96.65 / 91.94 98.13 / 94.27 97.34 / 93.18 98.68 / 95.41 hr → en 95.90 / 92.00 92.75 / 89.95 95.62 / 91.90 95.79 / 91.83 95.25 / 91.84 94.41 / 90.88 94.72 / 90.88 95.65 / 91.98 95.95 / 92.09 96.01 / 92.17 96.38 / 92.39 en → hu 96.34 / 93.18 93.98 / 89.63 97.45 / 94.80 97.53 / 95.16 95.50 / 93.29 89.96 / 84.71 92.18 / 87.67 95.51 / 92.03 96.96 / 94.42 97.08 / 93.95 98.29 / 95.89 hu → en 95.28 / 92.75 92.11 / 91.28 95.37 / 92.71 95.48 / 92.58 95.01 / 92.66 92.52 / 90.76 93.70 / 91.20 95.32 / 92.49 95.78 / 92.61 95.65 / 92.80 96.16 / 92.99 en → id 96.22 / 90.87 92.98 / 85.03 97.03 / 92.04 96.96 / 91.96 97.26 / 92.47 95.13 / 89.29 96.84 / 91.36 96.88 / 91.62 98.15 / 93.46 97.19 / 92.17 98.81 / 94.43 id → en 96.31 / 89.47 93.63 / 87.49 96.44 / 89.62 96.75 / 89.72 96.55 / 89.64 95.60 / 88.72 95.95 / 88.86 96.71 / 89.74 96.91 / 89.80 96.79 / 89.93 97.04 / 89.95 en → it 97.25 / 92.05 95.26 / 87.70 97.43 / 91.98 97.34 / 91.86 96.85 / 92.33 95.61 / 89.24 96.68 / 91.02 97.15 / 91.49 98.20 / 93.77 97.45 / 92.25 98.54 / 94.64 it → en 96.05 / 92.61 93.85 / 91.16 96.38 / 92.79 96.33 / 92.82 95.59 / 92.73 95.80 / 92.26 96.08 / 92.36 96.50 / 92.95 96.64 / 93.17 96.67 / 93.16 96.69 / 93.20 en → ja 94.32 / 95.29 80.62 / 88.79 94.67 / 95.37 94.98 / 95.58 93.87 / 95.30 90.62 / 91.59 93.19 / 93.31 94.29 / 94.56 95.79 / 95.49 95.19 / 95.13 96.39 / 96.05 ja → en 95.29 / 89.20 91.62 / 87.49 95.50 / 89.46 95.68 / 89.51 95.27 / 89.35 92.75 / 87.81 94.21 / 88.16 95.27 / 89.14 95.82 / 89.33 95.75 / 89.74 96.45 / 89.83 en → k 87.58 / 91.16 85.47 / 89.50 93.10 / 94.54 92.73 / 94.22 91.88 / 94.35 81.59 / 88.52 84.76 / 90.95 91.16 / 93.28 92.12 / 94.53 92.61 / 94.36 94.65 / 95.57 k → en 67.80 / 87.90 54.11 / 81.99 69.54 / 87.88 70.43 / 88.06 70.38 / 87.85 64.91 / 86.06 68.02 / 86.36 69.07 / 87.70 71.70 / 87.75 70.77 / 87.84 73.06 / 88.08 en → km 80.72 / 83.84 77.67 / 78.52 86.09 / 89.67 87.28 / 90.26 83.85 / 88.93 79.34 / 84.77 83.41 / 86.45 86.08 / 88.99 89.34 / 90.12 86.96 / 89.78 90.60 / 91.11 km → en 67.70 / 87.76 59.38 / 85.69 69.36 / 87.68 69.21 / 87.83 70.97 / 87.71 65.97 / 86.25 71.40 / 86.54 70.35 / 87.70 74.84 / 87.76 71.94 / 87.92 76.48 / 88.02 en → ko 96.11 / 92.10 89.66 / 89.11 96.52 / 92.60 96.57 / 92.76 95.54 / 92.77 91.31 / 87.95 94.13 / 90.00 95.47 / 91.49 97.17 / 92.91 96.27 / 92.27 97.85 / 93.73 ko → en 94.54 / 88.93 91.05 / 87.28 95.24 / 89.11 95.67 / 89.24 94.58 / 88.99 92.05 / 87.64 93.66 / 87.81 94.28 / 88.77 95.29 / 88.92 95.06 / 89.17 95.53 / 89.24 en → lo 76.90 / 76.37 80.79 / 79.47 87.72 / 87.39 88.64 / 87.87 85.04 / 85.50 79.35 / 81.05 83.75 / 83.31 88.03 / 86.61 91.44 / 87.68 88.63 / 87.35 92.68 / 88.82 lo → en 81.47 / 84.14 75.09 / 82.41 81.72 / 83.90 82.88 / 84.15 82.87 / 83.97 76.77 / 82.44 79.87 / 82.41 81.47 / 83.94 83.91 / 83.95 82.58 / 84.14 84.64 / 84.20 en → ms 91.77 / 84.36 90.79 / 81.46 95.63 / 89.42 95.58 / 89.38 95.82 / 90.45 94.14 / 87.66 95.93 / 89.76 95.53 / 89.36 97.54 / 91.87 95.81 / 89.65 98.12 / 92.87 ms → en 96.52 / 88.37 93.68 / 86.26 96.25 / 88.33 96.35 / 88.33 95.81 / 88.22 95.37 / 87.57 95.58 / 87.57 96.31 / 88.40 96.63 / 88.38 96.59 / 88.44 96.86 / 88.55 en → my 75.82 / 82.87 76.71 / 79.75 86.47 / 90.25 87.55 / 90.41 85.01 / 89.60 71.55 / 81.57 78.38 / 85.45 85.22 / 89.31 89.51 / 90.57 88.18 / 90.61 92.18 / 91.44 my → en 68.83 / 88.89 63.31 / 87.55 70.91 / 88.97 70.75 / 89.08 71.71 / 88.67 62.09 / 85.59 69.40 / 86.73 69.70 / 88.23 74.06 / 88.54 71.66 / 88.85 75.40 / 88.99 en → nb 97.74 / 93.62 94.17 / 87.10 98.27 / 94.57 98.40 / 94.83 97.81 / 95.04 95.78 / 90.47 97.04 / 92.17 97.74 / 93.78 98.29 / 94.87 98.20 / 94.69 98.81 / 95.91 nb → en 96.64 / 92.56 81.83 / 80.18 96.71 / 92.65 96.59 / 92.74 96.19 / 92.55 95.41 / 91.44 96.01 / 91.78 96.55 / 92.34 96.77 / 92.40 97.02 / 92.64 97.10 / 92.77 en → nl 98.18 / 92.56 96.20 / 87.60 98.14 / 92.54 98.15 / 92.68 98.05 / 92.85 96.03 / 88.08 96.79 / 89.06 97.86 / 91.49 98.49 / 92.83 98.24 / 92.61 98.92 / 94.03 nl → en 96.63 / 92.46 94.94 / 91.19 96.82 / 92.70 97.04 / 92.62 96.51 / 92.59 95.88 / 91.75 96.03 / 91.71 96.91 / 92.66 97.16 / 92.81 97.09 / 92.74 97.31 / 92.99 en → pl 97.13 / 91.55 93.18 / 85.15 97.26 / 91.24 97.20 / 91.51 97.12 / 91.77 91.62 / 82.28 93.32 / 85.34 96.30 / 89.70 97.39 / 92.28 97.29 / 91.31 98.38 / 93.90 pl → en 95.15 / 78.74 92.54 / 77.65 95.65 / 78.76 95.80 / 78.63 94.96 / 78.71 94.55 / 77.96 94.99 / 78.02 95.93 / 78.77 96.33 / 78.77 95.87 / 78.87 96.49 / 79.08 en → pt 97.32 / 91.62 95.58 / 86.99 97.52 / 91.64 97.50 / 91.54 97.28 / 92.20 96.46 / 90.07 97.20 / 91.77 97.37 / 91.67 98.38 / 93.79 97.89 / 92.57 98.80 / 94.51 pt → en 96.62 / 93.80 95.04 / 92.01 96.58 / 93.73 96.71 / 93.82 96.15 / 93.76 96.43 / 93.51 96.69 / 93.63 96.95 / 94.03 97.17 / 94.13 96.98 / 94.05 97.32 / 94.24 en → ro 96.54 / 94.67 93.03 / 89.51 97.45 / 95.82 97.42 / 95.91 96.76 / 96.43 93.38 / 90.88 94.47 / 92.85 96.32 / 94.55 97.51 / 96.47 96.98 / 95.70 98.16 / 97.31 ro → en 95.35 / 97.24 92.98 / 96.11 95.37 / 97.01 95.61 / 97.19 95.06 / 97.14 94.39 / 96.52 95.02 / 96.78 95.78 / 97.29 96.12 / 97.37 95.91 / 97.32 95.99 / 97.43 en → ru 96.38 / 91.63 93.65 / 85.43 96.54 / 92.04 96.69 / 92.10 96.12 / 91.63 91.54 / 84.79 93.60 / 87.21 95.47 / 89.93 96.69 / 91.96 96.20 / 91.39 97.47 / 93.38 ru → en 95.15 / 86.98 93.10 / 85.55 95.07 / 86.92 95.48 / 86.79 95.26 / 87.03 94.31 / 86.33 94.71 / 86.38 95.74 / 87.18 96.10 / 87.22 95.74 / 87.35 96.17 / 87.45 en → sk 95.65 / 91.66 92.43 / 87.78 96.33 / 92.70 96.46 / 92.98 96.53 / 93.38 90.96 / 85.15 93.05 / 88.56 95.25 / 90.83 97.02 / 93.58 96.24 / 92.29 97.80 / 95.00 sk → en 95.04 / 92.07 90.93 / 89.90 95.37 / 92.02 95.62 / 91.92 94.51 / 91.96 93.57 / 90.28 94.33 / 90.79 95.34 / 91.79 95.82 / 92.10 95.67 / 92.09 95.99 / 92.34 en → sl 96.32 / 91.35 92.30 / 87.08 97.08 / 92.57 97.37 / 92.93 96.05 / 93.02 91.00 / 83.30 93.23 / 86.57 95.52 / 90.17 96.96 / 92.78 96.75 / 92.03 98.13 / 94.68 sl → en 94.73 / 91.75 91.42 / 90.45 94.88 / 91.86 95.02 / 91.89 94.53 / 91.88 93.46 / 90.52 94.10 / 90.48 95.20 / 91.71 95.71 / 91.82 95.37 / 92.15 95.75 / 92.23 en → sv 97.77 / 93.28 96.15 / 89.82 98.39 / 94.17 98.61 / 94.63 97.73 / 94.69 95.87 / 89.83 96.60 / 90.80 97.71 / 93.23 98.50 / 94.80 98.37 / 94.30 98.96 / 95.65 sv → en 96.50 / 93.24 94.35 / 91.81 96.55 / 93.11 96.48 / 93.06 96.18 / 93.15 95.59 / 92.17 95.72 / 92.23 96.77 / 93.20 97.06 / 93.19 96.86 / 93.37 97.23 / 93.52 en → ta 83.16 / 83.20 79.74 / 77.14 87.81 / 85.98 89.68 / 87.06 86.62 / 85.15 76.42 / 77.54 82.52 / 80.87 85.71 / 84.33 90.52 / 86.18 87.29 / 85.59 92.27 / 87.61 ta → en 87.64 / 76.74 84.07 / 76.14 87.78 / 76.89 88.48 / 77.05 88.06 / 76.77 80.52 / 75.04 82.81 / 75.49 86.18 / 76.50 87.92 / 76.71 87.91 / 76.88 89.19 / 77.01 en → th 93.88 / 88.78 85.61 / 79.56 95.96 / 90.55 96.69 / 91.02 95.43 / 90.28 90.79 / 85.75 93.25 / 87.71 95.26 / 89.65 97.07 / 91.16 95.99 / 90.40 97.96 / 92.24 th → en 94.51 / 87.71 91.06 / 85.76 95.27 / 88.08 95.45 / 88.19 94.82 / 88.00 92.92 / 86.49 94.24 / 86.85 94.82 / 87.94 95.80 / 88.01 95.56 / 88.27 96.25 / 88.35 en → tl 85.61 / 80.84 85.48 / 80.45 91.90 / 87.64 92.32 / 88.06 91.59 / 88.46 86.67 / 82.54 90.40 / 85.08 90.30 / 86.41 93.87 / 89.39 91.39 / 87.57 95.22 / 90.72 tl → en 93.06 / 89.53 90.81 / 88.31 92.92 / 89.47 92.87 / 89.46 92.66 / 89.46 90.87 / 88.18 91.91 / 88.45 92.58 / 89.31 93.38 / 89.47 93.24 / 89.61 93.88 / 89.79 en → tr 94.33 / 92.16 89.51 / 85.55 94.75 / 91.95 94.85 / 91.87 94.12 / 92.13 88.94 / 85.64 91.99 / 89.00 93.05 / 90.10 96.28 / 93.62 94.74 / 91.81 97.08 / 94.55 tr → en 91.44 / 90.70 88.62 / 89.60 92.36 / 90.65 92.60 / 90.67 92.16 / 90.70 90.00 / 89.16 90.84 / 89.08 92.21 / 90.44 92.62 / 90.43 92.81 / 90.89 93.46 / 90.76 en → ur 83.31 / 79.27 81.91 / 78.28 93.09 / 86.17 93.74 / 86.91 91.32 / 85.96 81.88 / 77.62 87.69 / 81.63 90.03 / 84.36 94.85 / 86.92 92.20 / 86.01 96.01 / 88.31 ur → en 90.25 / 86.64 87.40 / 85.35 90.88 / 86.75 91.14 / 86.97 91.09 / 86.59 85.12 / 84.63 87.08 / 84.96 90.04 / 86.41 91.14 / 86.34 91.18 / 86.99 91.86 / 86.81 en → uz 82.85 / 87.03 83.81 / 87.17 89.84 / 91.86 90.48 / 91.88 89.02 / 91.75 78.49 / 84.61 83.39 / 87.96 87.87 / 90.31 91.02 / 92.73 89.54 / 91.48 92.91 / 93.61 uz → en 68.39 / 86.87 59.51 / 85.22 70.52 / 87.04 70.59 / 87.03 70.55 / 86.95 64.81 / 84.51 68.94 / 85.34 70.04 / 86.56 72.33 / 86.72 71.53 / 87.06 73.95 / 87.16 en → vi 95.50 / 92.07 89.92 / 84.13 95.75 / 92.77 96.17 / 93.06 95.81 / 92.93 93.18 / 89.97 95.21 / 91.74 95.02 / 91.96 97.18 / 93.85 95.79 / 92.62 97.72 / 94.37 vi → en 95.56 / 89.86 93.76 / 88.47 95.85 / 89.91 96.06 / 89.93 95.77 / 90.00 94.78 / 89.10 95.15 / 89.22 96.03 / 89.98 96.46 / 90.24 96.13 / 90.18 96.55 / 90.38 en → yue 82.57 / 80.31 76.47 / 74.48 88.83 / 86.64 88.34 / 86.25 89.23 / 87.32 85.86 / 83.54 86.92 / 84.68 89.36 / 86.65 91.68 / 88.50 90.37 / 87.45 92.66 / 89.24 yue → en 93.49 / 84.73 86.79 / 75.30 96.02 / 86.92 96.24 / 87.10 95.68 / 87.15 94.14 / 85.56 94.64 / 86.08 95.90 / 86.90 96.09 / 87.11 96.18 / 87.23 96.34 / 87.37 en → zhs 92.46 / 88.04 74.32 / 62.17 93.33 / 88.78 93.43 / 88.91 93.40 / 88.80 90.42 / 85.65 92.60 / 87.52 93.05 / 88.09 94.99 / 89.69 93.89 / 88.80 95.80 / 90.62 zhs → en 97.17 / 87.37 95.65 / 85.36 97.02 / 87.53 97.04 / 87.64 97.14 / 87.75 96.36 / 86.85 96.49 / 87.21 96.97 / 87.61 97.46 / 87.97 97.18 / 87.90 97.62 / 88.26 en → zht 92.07 / 88.64 53.15 / 38.63 93.28 / 89.77 93.27 / 89.86 93.23 / 89.56 89.19 / 86.03 92.03 / 87.82 92.36 / 88.74 94.93 / 90.49 93.45 / 89.36 95.75 / 91.24 zht → en 96.24 / 87.56 93.07 / 83.92 96.26 / 87.69 96.23 / 87.78 96.14 / 87.80 94.89 / 86.56 95.37 / 86.97 96.23 / 87.81 96.63 / 87.94 96.35 / 88.03 96.70 / 88.08 Table 9: Reference-free English-centric evaluation results (XCOMET / COMETKiwi) of baseline models (Part I) and MiLMMT models on the FLORES+ benchmark. Direction TranslateGemma-4B TranslateGemma-12B TranslateGemma-27B GemmaX2-28-2B GemmaX2-28-9B Hunyuan-MT-7B HY-MT1.5-1.8B HY-MT1.5-7B HY-MT2-1.8B HY-MT2-7B HY-MT2-30B-A3B Seed-X-Instruct-7B Seed-X-PPO-7B Tower-Plus-2B Tower-Plus-9B Tower-Plus-72B zhs → ar 78.48 / 73.15 81.62 / 76.79 83.13 / 77.26 77.98 / 71.25 82.40 / 74.60 83.67 / 76.35 80.80 / 73.82 84.28 / 77.25 81.46 / 73.69 83.97 / 76.58 83.39 / 75.80 78.17 / 71.29 82.11 / 75.08 - / - - / - - / - ar → zhs 68.08 / 68.66 70.55 / 70.69 71.60 / 71.19 68.83 / 70.57 71.75 / 72.40 73.04 / 71.40 70.80 / 69.76 73.87 / 71.00 72.49 / 70.83 74.49 / 71.46 74.36 / 72.82 65.91 / 70.45 65.42 / 69.60 - / - - / - - / - zhs → az 40.12 / 51.10 75.66 / 81.55 81.98 / 84.82 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - az → zhs 42.27 / 73.69 49.12 / 78.89 50.71 / 79.92 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → bg 88.55 / 81.80 92.12 / 85.72 92.94 / 86.44 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - bg → zhs 66.45 / 73.69 68.36 / 76.21 69.48 / 76.60 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → bn 84.89 / 72.44 87.52 / 76.05 89.18 / 76.30 84.72 / 73.63 88.10 / 75.71 87.97 / 74.38 84.46 / 71.06 89.20 / 77.23 85.64 / 72.10 90.41 / 75.98 90.58 / 76.10 - / - - / - - / - - / - - / - bn → zhs 58.79 / 70.51 62.46 / 73.46 63.78 / 74.14 61.59 / 74.57 64.42 / 76.75 61.73 / 70.15 63.92 / 70.39 64.77 / 70.57 64.98 / 71.01 67.08 / 72.70 68.69 / 76.20 - / - - / - - / - - / - - / - zhs → ca 82.24 / 76.08 86.99 / 80.71 88.26 / 81.97 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - ca → zhs 63.09 / 75.26 68.51 / 79.12 67.36 / 79.76 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → cs 75.97 / 83.84 80.60 / 88.03 83.05 / 88.77 80.85 / 84.58 85.01 / 87.48 84.27 / 88.01 78.02 / 82.47 84.40 / 88.18 81.81 / 83.99 84.78 / 87.61 84.51 / 87.83 76.31 / 84.12 83.44 / 87.93 79.58 / 83.64 83.94 / 87.73 84.49 / 87.40 cs → zhs 53.03 / 80.59 57.01 / 83.56 56.60 / 83.75 54.51 / 81.37 57.24 / 82.88 60.67 / 84.48 58.41 / 81.90 61.01 / 83.50 58.88 / 81.51 61.98 / 84.18 59.96 / 84.62 48.66 / 81.84 49.70 / 82.70 51.80 / 79.16 55.00 / 82.37 55.30 / 82.89 zhs → da 90.51 / 83.73 93.00 / 86.57 93.80 / 87.71 - / - - / - - / - - / - - / - - / - - / - - / - 90.59 / 83.59 93.23 / 86.57 91.29 / 83.50 93.36 / 85.69 93.15 / 85.29 da → zhs 75.25 / 81.93 78.48 / 84.18 78.97 / 84.75 - / - - / - - / - - / - - / - - / - - / - - / - 72.49 / 82.93 73.27 / 83.63 74.91 / 80.68 77.84 / 83.09 77.73 / 83.94 zhs → de 93.58 / 78.67 95.69 / 81.38 96.15 / 82.46 94.44 / 79.86 95.70 / 82.26 96.15 / 81.75 93.99 / 76.73 96.17 / 81.76 94.87 / 79.66 96.30 / 82.12 96.30 / 82.95 94.33 / 80.99 95.99 / 82.56 93.85 / 78.31 95.80 / 82.39 95.91 / 82.30 de → zhs 78.15 / 82.57 82.16 / 85.62 81.83 / 86.05 81.13 / 82.21 82.44 / 83.17 83.74 / 87.67 81.20 / 85.52 83.07 / 87.28 82.34 / 84.14 83.48 / 86.34 83.39 / 86.45 77.35 / 82.49 76.07 / 84.21 78.45 / 80.33 81.78 / 83.29 81.66 / 83.53 zhs → el 80.70 / 71.70 86.61 / 76.82 87.38 / 77.67 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - el → zhs 57.34 / 62.73 60.61 / 65.49 61.18 / 65.91 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → en 96.37 / 86.48 97.02 / 87.03 97.34 / 87.76 97.01 / 87.46 97.18 / 87.85 97.26 / 87.51 96.43 / 85.71 97.15 / 87.42 96.73 / 87.09 97.46 / 87.64 97.62 / 87.92 96.63 / 87.03 97.07 / 87.65 96.54 / 87.16 97.08 / 87.69 97.28 / 87.79 en → zhs 91.98 / 87.14 94.05 / 88.92 94.45 / 89.86 92.50 / 87.66 93.21 / 88.48 95.56 / 89.93 93.68 / 88.42 95.04 / 89.84 93.08 / 88.03 95.06 / 89.91 95.11 / 90.01 93.41 / 88.63 94.31 / 89.93 90.60 / 85.96 92.33 / 88.05 93.36 / 88.95 zhs → es 89.13 / 79.18 90.31 / 80.42 90.89 / 80.66 90.48 / 79.64 91.57 / 80.95 91.55 / 79.99 90.74 / 78.09 91.82 / 80.06 91.14 / 78.80 92.80 / 80.55 92.28 / 80.88 89.53 / 80.04 91.84 / 81.10 89.53 / 78.45 91.42 / 80.80 91.70 / 81.48 es → zhs 82.86 / 79.89 85.39 / 82.74 85.74 / 83.47 83.12 / 78.17 84.94 / 80.11 87.32 / 84.86 85.88 / 82.11 87.53 / 85.02 85.75 / 80.53 87.58 / 83.75 87.28 / 83.13 79.39 / 77.85 80.55 / 81.20 80.84 / 76.66 83.31 / 78.69 84.50 / 79.71 zhs → fa 83.93 / 78.59 88.99 / 82.95 89.47 / 83.78 85.47 / 79.52 88.28 / 81.88 86.32 / 80.00 82.69 / 77.31 88.00 / 82.03 84.82 / 78.52 89.49 / 82.72 89.37 / 82.61 - / - - / - - / - - / - - / - fa → zhs 73.15 / 73.14 77.01 / 75.97 78.52 / 77.11 75.31 / 74.34 77.38 / 76.38 78.11 / 76.38 75.41 / 73.26 79.96 / 76.59 76.82 / 73.74 81.05 / 77.23 80.69 / 77.84 - / - - / - - / - - / - - / - zhs → fi 78.25 / 84.10 86.43 / 89.55 87.94 / 91.28 - / - - / - - / - - / - - / - - / - - / - - / - 81.13 / 85.82 86.27 / 88.95 81.84 / 84.66 87.10 / 89.19 85.37 / 87.68 fi → zhs 51.78 / 81.33 58.78 / 83.94 58.76 / 84.36 - / - - / - - / - - / - - / - - / - - / - - / - 48.74 / 83.40 50.42 / 83.44 53.33 / 80.88 54.77 / 82.73 57.60 / 84.19 zhs → fr 77.61 / 78.80 79.38 / 81.08 80.52 / 81.24 81.78 / 80.76 83.88 / 82.45 81.91 / 81.35 80.34 / 78.52 82.34 / 81.06 82.18 / 80.26 83.52 / 82.60 83.58 / 82.37 79.28 / 81.08 82.39 / 81.99 79.16 / 79.26 82.68 / 81.99 83.27 / 82.58 fr → zhs 66.95 / 77.46 69.83 / 80.86 70.73 / 82.41 65.94 / 74.34 67.58 / 76.05 74.52 / 83.88 71.65 / 80.80 75.00 / 84.23 69.36 / 77.94 73.90 / 82.16 72.38 / 80.62 61.52 / 75.43 63.74 / 79.59 62.97 / 72.48 66.90 / 76.02 67.00 / 76.75 zhs → he 74.10 / 69.04 83.12 / 77.32 84.16 / 78.19 79.53 / 73.07 83.94 / 77.15 76.67 / 71.23 80.78 / 63.21 78.94 / 73.19 78.62 / 70.90 85.83 / 77.89 84.86 / 77.41 - / - - / - - / - - / - - / - he → zhs 70.93 / 71.50 75.31 / 74.22 76.38 / 75.01 74.17 / 73.60 75.71 / 74.82 75.40 / 74.20 74.15 / 71.48 76.22 / 73.25 75.86 / 72.04 79.27 / 74.88 79.87 / 76.09 - / - - / - - / - - / - - / - zhs → hi 84.60 / 53.88 88.80 / 57.82 89.43 / 58.29 83.18 / 54.73 87.24 / 57.70 88.80 / 55.66 84.83 / 51.43 88.89 / 56.84 85.66 / 52.84 90.13 / 55.50 90.38 / 55.94 - / - - / - 79.37 / 52.67 85.86 / 57.00 84.34 / 55.40 hi → zhs 71.54 / 55.12 75.37 / 57.28 76.28 / 57.98 72.57 / 58.92 75.12 / 60.70 76.92 / 58.42 74.88 / 56.24 78.69 / 57.57 75.29 / 57.85 79.76 / 59.03 79.44 / 59.47 - / - - / - 70.46 / 57.49 74.87 / 59.62 75.05 / 60.45 zhs → hr 81.70 / 77.91 88.45 / 84.46 90.19 / 85.81 - / - - / - - / - - / - - / - - / - - / - - / - 80.77 / 77.19 87.13 / 82.18 - / - - / - - / - hr → zhs 67.23 / 79.01 71.71 / 82.29 72.18 / 82.93 - / - - / - - / - - / - - / - - / - - / - - / - 62.73 / 78.66 64.43 / 79.84 - / - - / - - / - zhs → hu 79.09 / 77.60 88.93 / 86.76 91.90 / 88.62 - / - - / - - / - - / - - / - - / - - / - - / - 88.10 / 84.60 91.45 / 87.97 87.74 / 84.32 91.74 / 87.78 90.91 / 86.70 hu → zhs 65.18 / 79.64 71.69 / 83.51 71.33 / 83.49 - / - - / - - / - - / - - / - - / - - / - - / - 62.21 / 82.35 64.01 / 82.41 65.39 / 79.98 68.83 / 82.97 69.76 / 83.68 zhs → id 92.29 / 83.07 93.64 / 84.65 93.80 / 84.81 91.76 / 81.65 92.69 / 82.75 93.91 / 84.70 92.54 / 83.22 93.98 / 85.50 92.66 / 83.13 94.40 / 85.20 93.64 / 84.03 89.71 / 79.00 92.20 / 82.25 - / - - / - - / - id → zhs 82.37 / 80.40 85.22 / 82.16 84.42 / 82.44 83.97 / 81.99 85.41 / 83.47 85.17 / 81.97 83.24 / 80.39 85.37 / 81.83 84.92 / 81.97 86.36 / 82.71 86.40 / 83.71 80.49 / 81.49 79.27 / 80.42 - / - - / - - / - zhs → it 84.55 / 79.06 87.04 / 80.75 87.86 / 81.57 86.92 / 79.49 88.78 / 80.72 88.67 / 81.26 87.40 / 78.99 88.67 / 81.63 88.43 / 79.18 89.70 / 81.77 89.31 / 81.85 85.59 / 79.38 88.78 / 81.63 85.15 / 78.24 88.17 / 80.91 88.66 / 81.55 it → zhs 75.41 / 79.52 79.12 / 82.49 78.49 / 83.20 76.56 / 78.75 77.49 / 80.63 81.27 / 84.92 79.14 / 82.38 81.26 / 84.82 78.38 / 80.25 82.07 / 83.99 81.28 / 83.29 69.01 / 77.87 71.00 / 80.72 73.11 / 76.63 75.48 / 79.65 76.61 / 79.54 zhs → ja 84.28 / 88.66 86.46 / 89.86 86.30 / 90.17 88.20 / 90.02 88.97 / 90.69 89.31 / 90.75 88.85 / 89.86 89.07 / 90.78 88.71 / 90.15 89.10 / 90.67 89.64 / 90.81 82.47 / 87.01 87.50 / 89.51 86.94 / 89.55 88.12 / 90.32 88.24 / 90.37 ja → zhs 63.54 / 82.94 66.80 / 84.79 67.69 / 85.14 65.27 / 82.77 66.40 / 83.56 69.27 / 85.87 67.94 / 84.38 70.41 / 85.55 68.69 / 83.74 70.51 / 85.25 70.52 / 84.63 59.23 / 82.18 60.20 / 83.61 63.54 / 81.56 65.35 / 83.24 66.15 / 83.61 zhs → k 41.48 / 63.39 65.69 / 85.29 74.19 / 88.71 - / - - / - 29.33 / 62.61 26.81 / 54.22 28.55 / 62.22 26.62 / 55.81 26.37 / 63.68 32.91 / 66.53 - / - - / - - / - - / - - / - k → zhs 34.02 / 74.55 40.01 / 79.91 39.96 / 80.07 - / - - / - 28.56 / 62.80 19.89 / 49.25 29.59 / 63.48 20.46 / 49.36 35.84 / 71.73 42.62 / 80.15 - / - - / - - / - - / - - / - zhs → km 31.07 / 49.36 62.02 / 78.61 71.09 / 82.67 73.18 / 82.60 77.67 / 84.86 64.37 / 80.58 97.49 / 50.65 68.37 / 81.95 75.57 / 83.64 79.79 / 85.49 79.37 / 85.90 - / - - / - - / - - / - - / - km → zhs 36.14 / 79.07 40.98 / 82.73 42.95 / 83.58 41.01 / 82.31 43.42 / 83.90 42.06 / 80.71 43.91 / 81.17 43.20 / 81.31 45.52 / 82.01 47.17 / 83.31 46.08 / 84.34 - / - - / - - / - - / - - / - zhs → ko 90.67 / 89.13 93.32 / 91.00 94.56 / 91.34 91.97 / 89.58 93.61 / 90.91 94.58 / 91.53 93.00 / 88.63 94.47 / 91.46 93.19 / 90.64 94.62 / 91.60 94.63 / 91.46 85.23 / 84.74 89.82 / 88.54 91.04 / 89.17 93.96 / 91.08 93.51 / 90.56 ko → zhs 66.75 / 84.75 70.74 / 86.77 70.39 / 87.20 69.11 / 86.40 71.32 / 87.60 71.43 / 87.01 70.25 / 85.59 72.63 / 86.90 72.54 / 86.67 73.82 / 87.54 74.02 / 87.89 62.66 / 85.22 62.44 / 85.66 66.66 / 85.06 69.39 / 87.19 71.01 / 87.70 zhs → lo 38.00 / 49.43 68.37 / 76.74 76.03 / 80.80 74.90 / 78.93 80.14 / 82.01 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - lo → zhs 43.39 / 74.72 49.08 / 79.36 51.23 / 80.16 51.14 / 78.95 54.45 / 81.10 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → ms 87.32 / 79.21 91.51 / 83.29 92.01 / 83.66 90.42 / 81.34 91.25 / 82.15 91.83 / 83.62 92.07 / 78.87 92.64 / 84.69 90.98 / 82.25 93.35 / 84.10 92.70 / 84.00 87.50 / 77.48 90.19 / 81.28 - / - - / - - / - ms → zhs 75.02 / 77.20 78.96 / 80.24 79.46 / 80.77 80.17 / 80.75 81.45 / 82.07 80.30 / 80.35 78.07 / 77.96 80.44 / 79.65 80.39 / 79.86 82.34 / 80.59 82.67 / 82.16 74.75 / 79.82 72.17 / 79.06 - / - - / - - / - zhs → my 35.33 / 58.33 54.51 / 77.39 66.61 / 82.77 68.04 / 82.83 76.30 / 86.16 60.48 / 80.44 73.29 / 79.00 64.91 / 82.75 73.51 / 85.22 77.58 / 86.83 78.60 / 87.20 - / - - / - - / - - / - - / - my → zhs 23.50 / 78.68 25.59 / 82.81 26.16 / 83.96 25.91 / 82.08 27.35 / 84.73 24.85 / 80.65 26.55 / 81.91 26.10 / 82.37 28.29 / 83.25 28.86 / 84.71 28.23 / 85.44 - / - - / - - / - - / - - / - zhs → nb 91.63 / 84.80 93.55 / 87.54 94.08 / 88.16 - / - - / - - / - - / - - / - - / - - / - - / - 87.78 / 81.01 91.51 / 84.92 91.04 / 83.68 92.91 / 85.54 92.95 / 86.03 nb → zhs 74.89 / 79.68 78.83 / 82.46 78.13 / 82.54 - / - - / - - / - - / - - / - - / - - / - - / - 71.95 / 80.36 71.19 / 80.84 74.07 / 78.57 77.42 / 81.74 77.97 / 81.87 zhs → nl 89.68 / 81.97 92.10 / 84.61 93.41 / 85.46 91.22 / 82.09 93.05 / 84.35 91.41 / 82.23 89.77 / 79.49 91.61 / 82.92 91.09 / 80.57 93.64 / 84.44 94.28 / 85.30 89.73 / 80.94 92.41 / 83.94 90.23 / 81.63 92.31 / 84.25 92.36 / 84.33 nl → zhs 74.01 / 83.24 77.56 / 86.08 78.33 / 86.69 76.68 / 83.19 78.05 / 84.96 79.94 / 87.48 77.98 / 85.08 80.19 / 87.36 79.16 / 84.46 81.53 / 87.21 80.98 / 87.44 69.86 / 83.64 71.88 / 85.28 72.07 / 81.71 75.69 / 84.05 76.16 / 84.59 zhs → pl 84.27 / 78.26 87.90 / 81.68 88.66 / 82.57 88.05 / 79.42 89.74 / 81.56 84.11 / 78.12 83.07 / 75.21 87.36 / 79.16 86.15 / 77.40 89.73 / 80.88 90.18 / 81.73 83.62 / 77.24 88.48 / 80.65 85.74 / 78.17 89.29 / 81.21 89.10 / 80.80 pl → zhs 61.55 / 71.69 64.22 / 73.64 64.45 / 74.25 61.01 / 70.48 62.59 / 72.33 66.37 / 76.03 64.91 / 72.64 66.38 / 75.60 64.05 / 71.31 68.31 / 75.40 67.83 / 75.32 54.65 / 71.09 55.09 / 72.05 59.31 / 69.85 62.06 / 72.21 62.81 / 72.95 zhs → pt 89.88 / 78.54 90.59 / 80.40 91.41 / 80.68 90.52 / 78.98 92.01 / 80.22 91.97 / 80.16 91.37 / 78.11 92.42 / 80.64 90.93 / 78.43 92.40 / 80.49 92.59 / 80.38 88.31 / 77.79 91.01 / 80.42 88.96 / 77.67 90.59 / 79.80 91.02 / 80.34 pt → zhs 82.79 / 79.57 85.18 / 82.25 85.29 / 83.11 83.50 / 78.56 84.57 / 80.29 87.20 / 84.12 84.95 / 82.19 86.86 / 84.55 85.26 / 80.98 87.73 / 83.68 87.84 / 83.87 81.12 / 79.06 81.14 / 80.74 81.31 / 77.64 83.77 / 79.21 83.95 / 80.19 zhs → ro 80.90 / 82.20 83.77 / 84.74 85.40 / 86.17 - / - - / - - / - - / - - / - - / - - / - - / - 78.33 / 80.77 84.81 / 85.06 83.07 / 82.56 86.60 / 85.65 85.73 / 85.15 ro → zhs 59.32 / 73.86 62.63 / 76.39 63.73 / 77.05 - / - - / - - / - - / - - / - - / - - / - - / - 54.27 / 74.01 56.14 / 74.41 56.30 / 71.68 58.92 / 74.41 60.02 / 75.46 zhs → ru 90.62 / 77.03 92.86 / 78.92 92.89 / 78.73 90.43 / 77.09 92.64 / 79.54 92.84 / 78.52 90.58 / 76.09 92.55 / 78.65 90.86 / 76.96 93.22 / 79.40 93.67 / 79.76 91.73 / 78.12 93.21 / 79.51 90.58 / 76.86 92.77 / 79.59 92.99 / 79.61 ru → zhs 80.81 / 73.28 82.93 / 76.02 83.50 / 77.11 80.97 / 71.97 82.20 / 73.56 85.21 / 79.13 83.87 / 77.21 85.36 / 78.48 83.80 / 75.88 86.08 / 78.38 85.60 / 77.82 78.30 / 71.71 79.52 / 74.33 79.71 / 71.19 81.01 / 73.20 81.22 / 73.55 zhs → sk 75.70 / 80.03 83.42 / 86.38 85.43 / 87.82 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - sk → zhs 57.17 / 79.28 62.17 / 82.56 62.59 / 83.00 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → sl 69.05 / 71.71 81.74 / 82.74 83.81 / 85.33 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - sl → zhs 57.32 / 76.48 61.89 / 80.84 61.28 / 81.39 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → sv 90.95 / 83.91 93.71 / 87.36 94.46 / 88.03 - / - - / - - / - - / - - / - - / - - / - - / - 90.72 / 83.89 93.40 / 86.70 91.76 / 84.33 93.89 / 86.72 93.33 / 86.01 sv → zhs 76.76 / 82.76 81.07 / 85.52 80.58 / 85.57 - / - - / - - / - - / - - / - - / - - / - - / - 75.16 / 84.00 75.04 / 84.35 76.54 / 81.41 79.55 / 84.62 79.66 / 84.82 zhs → ta 78.58 / 71.24 83.78 / 76.71 84.57 / 76.60 - / - - / - 72.66 / 66.72 75.38 / 67.12 75.57 / 72.11 78.76 / 70.28 85.70 / 74.91 85.70 / 75.72 - / - - / - - / - - / - - / - ta → zhs 47.19 / 65.46 52.20 / 69.57 53.14 / 69.95 - / - - / - 51.39 / 65.69 51.30 / 65.70 54.04 / 66.47 54.03 / 66.87 58.78 / 68.94 56.03 / 72.57 - / - - / - - / - - / - - / - zhs → th 81.24 / 78.97 86.56 / 83.99 87.66 / 84.29 83.47 / 80.71 85.94 / 83.28 86.40 / 84.13 83.92 / 78.26 87.62 / 84.99 84.92 / 81.74 89.20 / 84.91 88.60 / 84.68 80.86 / 79.04 84.67 / 81.73 - / - - / - - / - th → zhs 72.06 / 79.19 76.15 / 81.07 76.84 / 82.04 76.40 / 81.96 77.71 / 83.15 75.92 / 80.95 75.94 / 79.35 77.60 / 80.66 77.46 / 80.40 79.56 / 81.12 79.89 / 83.37 70.03 / 81.16 69.87 / 80.96 - / - - / - - / - zhs → tl 83.74 / 80.49 85.79 / 82.99 86.73 / 83.13 83.84 / 81.20 85.45 / 82.35 78.37 / 77.77 82.40 / 79.24 80.63 / 79.87 84.17 / 80.95 86.87 / 82.73 86.67 / 82.87 - / - - / - - / - - / - - / - tl → zhs 45.58 / 76.68 48.68 / 79.80 48.77 / 79.90 49.84 / 79.13 51.07 / 80.59 50.13 / 78.96 47.77 / 76.82 50.08 / 78.28 50.39 / 77.96 51.48 / 79.92 52.20 / 81.44 - / - - / - - / - - / - - / - zhs → tr 80.20 / 82.23 83.35 / 85.53 85.19 / 86.40 82.53 / 82.17 84.39 / 84.73 86.31 / 86.28 83.92 / 84.47 86.38 / 86.23 83.27 / 83.49 86.37 / 85.87 86.76 / 86.36 77.60 / 80.29 82.82 / 83.68 - / - - / - - / - tr → zhs 58.89 / 78.01 64.01 / 81.16 63.87 / 81.17 61.82 / 80.02 64.07 / 81.71 66.47 / 81.37 64.45 / 79.64 67.84 / 81.41 66.19 / 79.75 68.73 / 81.67 68.02 / 82.62 55.75 / 80.45 56.07 / 79.88 - / - - / - - / - zhs → ur 70.48 / 65.88 84.35 / 76.88 86.94 / 78.97 81.36 / 74.31 86.41 / 77.25 80.08 / 71.85 79.27 / 69.63 82.42 / 74.91 81.99 / 73.35 88.40 / 78.29 88.21 / 78.44 - / - - / - - / - - / - - / - ur → zhs 52.24 / 66.84 57.32 / 71.21 58.67 / 71.34 55.77 / 71.68 58.64 / 73.86 59.24 / 68.51 55.33 / 66.51 61.21 / 68.74 57.47 / 68.04 63.54 / 70.56 63.58 / 73.53 - / - - / - - / - - / - - / - zhs → uz 18.84 / 26.75 66.01 / 83.52 72.80 / 86.50 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - uz → zhs 25.76 / 74.29 30.19 / 79.90 31.13 / 80.96 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → vi 90.72 / 84.66 92.43 / 86.73 93.27 / 87.05 91.34 / 85.22 92.11 / 85.80 93.41 / 87.52 91.83 / 85.70 93.27 / 87.65 92.06 / 85.68 93.97 / 87.09 93.82 / 87.27 87.88 / 81.62 91.30 / 85.14 - / - - / - - / - vi → zhs 80.85 / 81.57 84.65 / 83.72 85.27 / 84.42 83.30 / 83.73 85.08 / 84.94 85.52 / 83.60 84.81 / 82.27 86.13 / 83.39 85.20 / 83.28 86.86 / 84.02 86.96 / 85.28 79.26 / 83.06 79.45 / 83.17 - / - - / - - / - zhs → yue 96.85 / 54.26 97.28 / 67.92 97.46 / 63.04 - / - - / - 94.48 / 43.01 95.20 / 52.97 94.23 / 55.54 92.24 / 48.68 93.84 / 63.82 92.80 / 38.87 - / - - / - - / - - / - - / - yue → zhs 96.21 / 71.91 95.81 / 82.78 96.07 / 83.70 - / - - / - 96.73 / 84.92 95.63 / 83.43 96.97 / 85.02 94.94 / 70.62 95.27 / 69.79 95.17 / 67.79 - / - - / - - / - - / - - / - zhs → zht 97.35 / 44.57 97.36 / 54.79 97.50 / 65.20 - / - - / - 97.26 / 74.45 96.77 / 74.01 97.44 / 71.70 97.04 / 59.17 97.74 / 62.98 96.62 / 24.65 - / - - / - 96.34 / 43.60 96.95 / 28.20 96.59 / 26.78 zht → zhs 96.82 / 53.89 95.83 / 74.53 96.00 / 75.58 - / - - / - 96.78 / 78.14 95.83 / 78.17 96.64 / 78.35 94.65 / 48.06 95.31 / 48.53 93.24 / 28.19 - / - - / - 94.44 / 61.51 94.78 / 55.37 94.59 / 49.22 Table 10: Reference-free Chinese-centric evaluation results (XCOMET / COMETKiwi) of baseline models (Part I) on the FLORES+ benchmark. Direction Google Translate NLLB-54.5B Gemini 2.5 Pro Gemini 3 Pro GPT-5 MiLMMT-46-1B-v0.1 MiLMMT-46-1B-v1.0 MiLMMT-46-4B-v0.1 MiLMMT-46-4B-v1.0 MiLMMT-46-12B-v0.1 MiLMMT-46-12B-v1.0 zhs → ar 79.53 / 73.09 79.30 / 71.84 82.51 / 74.67 83.14 / 74.76 83.06 / 75.46 74.88 / 68.37 77.85 / 71.85 81.05 / 73.92 83.96 / 76.94 83.03 / 75.40 84.86 / 78.05 ar → zhs 66.40 / 70.69 52.06 / 58.70 70.50 / 72.42 71.81 / 72.89 71.76 / 73.05 67.78 / 68.55 71.52 / 70.30 70.30 / 71.67 73.99 / 72.37 71.87 / 73.29 75.00 / 73.59 zhs → az 76.37 / 79.72 74.45 / 76.04 84.65 / 85.37 84.75 / 85.69 85.47 / 86.09 72.42 / 77.31 76.26 / 80.69 82.79 / 84.23 84.44 / 86.17 84.49 / 85.74 86.96 / 87.57 az → zhs 42.77 / 79.66 35.20 / 69.50 51.11 / 81.48 51.15 / 82.01 53.65 / 82.06 44.88 / 75.64 50.64 / 77.61 49.20 / 80.64 54.71 / 80.90 50.47 / 81.51 57.09 / 81.96 zhs → bg 90.64 / 82.96 87.00 / 78.52 92.39 / 85.37 92.66 / 85.35 93.18 / 86.06 86.60 / 78.50 88.93 / 81.56 92.11 / 84.75 93.20 / 86.45 92.86 / 85.62 93.85 / 87.51 bg → zhs 63.15 / 73.91 41.99 / 52.23 67.76 / 76.00 67.38 / 75.83 68.75 / 76.29 64.10 / 72.59 69.84 / 74.82 67.40 / 75.25 72.61 / 77.68 68.25 / 76.39 73.58 / 78.85 zhs → bn 83.30 / 70.64 80.46 / 67.99 87.97 / 75.47 88.97 / 76.23 89.13 / 76.45 79.75 / 69.11 85.07 / 74.04 87.50 / 75.32 90.12 / 78.07 88.52 / 76.34 91.34 / 78.99 bn → zhs 59.53 / 74.29 44.52 / 62.58 63.45 / 76.97 64.11 / 77.01 66.13 / 76.99 57.77 / 71.90 63.58 / 73.08 63.27 / 75.41 66.63 / 75.81 64.41 / 77.06 68.68 / 76.81 zhs → ca 84.07 / 78.03 80.90 / 74.83 86.85 / 80.32 87.58 / 80.52 88.66 / 81.41 83.82 / 76.11 86.97 / 79.47 88.01 / 80.52 89.23 / 82.30 88.46 / 81.14 90.52 / 83.46 ca → zhs 63.44 / 76.00 48.24 / 58.39 68.24 / 79.11 68.36 / 77.90 70.48 / 78.63 64.11 / 73.42 68.62 / 77.47 67.46 / 76.68 72.75 / 81.12 68.94 / 78.57 73.97 / 82.41 zhs → cs 81.01 / 85.66 75.05 / 80.70 84.32 / 87.73 85.25 / 88.25 85.86 / 88.66 75.07 / 79.51 78.98 / 83.85 84.87 / 87.06 84.94 / 88.82 85.68 / 88.18 85.49 / 89.68 cs → zhs 51.19 / 81.22 32.25 / 57.69 56.04 / 83.44 56.43 / 83.36 59.19 / 83.78 52.62 / 78.43 57.87 / 80.90 56.18 / 82.45 60.88 / 84.63 57.03 / 83.37 62.83 / 85.64 zhs → da 91.60 / 84.99 89.12 / 81.43 93.79 / 86.68 94.30 / 87.05 94.69 / 87.76 88.95 / 80.60 91.01 / 82.86 93.30 / 85.80 93.83 / 87.49 94.23 / 86.85 95.46 / 88.70 da → zhs 75.73 / 83.06 45.99 / 54.52 78.97 / 84.62 78.49 / 84.10 81.13 / 84.96 75.43 / 80.93 79.84 / 83.24 77.97 / 83.48 81.72 / 85.70 79.62 / 84.27 83.63 / 86.74 zhs → de 95.06 / 81.56 91.19 / 77.99 95.98 / 83.00 95.97 / 83.11 96.05 / 83.15 92.12 / 75.59 93.72 / 77.56 95.45 / 81.06 96.57 / 82.38 96.10 / 82.89 96.75 / 83.29 de → zhs 79.17 / 81.82 58.12 / 58.51 82.37 / 84.28 82.13 / 83.64 82.90 / 84.99 78.96 / 80.46 82.05 / 84.40 81.41 / 83.08 84.12 / 86.96 82.88 / 84.12 84.90 / 88.15 zhs → el 85.76 / 74.42 83.40 / 71.31 88.12 / 76.97 88.82 / 77.25 88.82 / 77.81 79.06 / 70.31 81.37 / 72.73 87.32 / 76.83 88.31 / 78.33 88.37 / 77.60 89.58 / 79.20 el → zhs 54.66 / 63.81 41.15 / 52.10 59.02 / 66.32 58.01 / 65.43 60.91 / 66.06 56.00 / 61.42 62.05 / 64.67 58.78 / 65.55 64.34 / 67.64 59.90 / 66.49 66.50 / 69.22 zhs → en 97.17 / 87.37 95.65 / 85.36 97.02 / 87.53 97.04 / 87.64 97.14 / 87.75 96.36 / 86.85 96.48 / 87.21 96.97 / 87.61 97.47 / 87.95 97.18 / 87.90 97.61 / 88.27 en → zhs 92.46 / 88.04 74.32 / 62.17 93.33 / 88.78 93.43 / 88.91 93.40 / 88.80 90.42 / 85.65 92.61 / 87.55 93.05 / 88.09 94.99 / 89.67 93.89 / 88.80 95.79 / 90.58 zhs → es 90.02 / 80.34 87.51 / 77.80 91.15 / 81.03 91.55 / 81.25 91.25 / 80.89 89.29 / 78.24 90.96 / 79.14 91.70 / 80.69 92.56 / 80.89 92.00 / 81.26 92.95 / 81.66 es → zhs 81.29 / 77.19 53.43 / 47.01 85.04 / 80.28 85.40 / 79.66 85.98 / 80.98 81.61 / 75.98 85.44 / 81.65 83.87 / 79.51 87.32 / 84.32 85.02 / 80.44 88.46 / 85.16 zhs → fa 86.07 / 79.52 82.13 / 76.25 89.03 / 81.86 88.99 / 82.00 89.02 / 82.51 78.17 / 73.43 82.89 / 77.46 87.22 / 80.76 89.18 / 83.11 88.66 / 81.86 91.37 / 85.01 fa → zhs 71.89 / 73.85 55.41 / 60.70 76.80 / 76.63 76.59 / 76.22 77.87 / 77.22 71.11 / 71.31 75.95 / 74.34 76.36 / 75.17 79.91 / 77.65 77.63 / 77.05 81.76 / 79.17 zhs → fi 84.70 / 88.28 77.46 / 81.62 89.25 / 90.52 89.73 / 90.73 89.97 / 91.13 74.15 / 77.15 78.83 / 80.99 87.12 / 88.11 88.30 / 90.20 88.72 / 90.02 90.98 / 91.90 fi → zhs 52.32 / 83.01 32.23 / 63.73 58.05 / 84.62 58.37 / 84.76 60.52 / 84.73 52.92 / 79.30 58.08 / 80.54 58.01 / 83.80 60.81 / 84.57 57.99 / 84.81 64.79 / 85.97 zhs → fr 80.48 / 81.63 76.83 / 77.16 81.98 / 82.61 82.96 / 82.66 82.35 / 82.25 78.94 / 77.20 80.93 / 79.08 83.04 / 82.05 83.19 / 81.94 83.72 / 82.68 83.82 / 82.49 fr → zhs 62.20 / 72.88 40.52 / 46.44 67.53 / 77.24 66.92 / 75.77 70.02 / 76.72 64.44 / 71.62 71.16 / 78.29 67.30 / 75.61 74.32 / 82.51 68.68 / 76.60 76.01 / 84.48 zhs → he 79.86 / 73.53 78.83 / 71.00 83.63 / 76.94 83.95 / 77.51 84.89 / 77.59 71.88 / 66.49 74.79 / 70.49 83.00 / 75.87 85.22 / 78.47 84.78 / 77.38 87.13 / 79.92 he → zhs 72.41 / 73.67 48.52 / 54.19 75.85 / 75.77 76.04 / 75.84 76.96 / 76.06 71.41 / 70.67 75.36 / 72.69 74.57 / 74.52 78.17 / 75.84 75.54 / 75.44 78.91 / 76.80 zhs → hi 86.20 / 52.91 81.39 / 51.81 88.23 / 56.06 88.59 / 56.93 89.32 / 57.34 78.31 / 52.18 83.85 / 54.96 86.66 / 56.19 89.50 / 58.25 88.43 / 57.86 91.19 / 59.01 hi → zhs 71.42 / 57.93 58.76 / 48.40 75.18 / 61.56 75.36 / 61.79 75.67 / 61.51 68.88 / 56.33 74.07 / 58.12 74.25 / 59.82 78.86 / 60.07 75.22 / 61.15 79.51 / 60.79 zhs → hr 87.10 / 82.00 83.56 / 79.18 90.02 / 84.76 90.20 / 85.05 90.71 / 85.21 82.94 / 76.90 86.32 / 80.53 89.24 / 83.54 90.90 / 85.51 90.18 / 84.72 91.57 / 86.50 hr → zhs 67.89 / 79.76 41.56 / 54.48 71.97 / 82.14 71.57 / 81.45 73.32 / 82.28 67.18 / 77.14 72.05 / 79.96 71.29 / 80.92 74.91 / 83.47 71.81 / 81.65 77.51 / 84.77 zhs → hu 89.67 / 86.70 84.83 / 80.70 93.27 / 89.23 93.52 / 89.41 93.85 / 89.83 82.15 / 78.58 85.35 / 81.98 90.99 / 86.59 92.07 / 88.65 93.04 / 88.81 93.96 / 90.40 hu → zhs 64.55 / 82.03 41.53 / 62.34 71.26 / 84.09 70.98 / 84.16 72.94 / 84.40 64.76 / 77.88 71.61 / 80.54 70.09 / 83.19 74.77 / 84.28 71.54 / 84.26 77.25 / 85.68 zhs → id 90.82 / 79.63 87.87 / 77.74 92.49 / 82.56 92.57 / 82.49 93.16 / 83.47 89.70 / 79.31 92.03 / 82.81 92.25 / 82.30 94.29 / 85.53 92.84 / 83.05 95.17 / 86.62 id → zhs 82.08 / 82.30 60.62 / 62.94 85.01 / 83.69 85.30 / 83.46 86.00 / 83.76 81.79 / 80.44 84.42 / 81.41 84.13 / 83.04 85.79 / 83.15 86.06 / 83.74 87.76 / 84.02 zhs → it 86.44 / 80.50 81.62 / 75.46 88.57 / 81.51 88.69 / 81.72 89.19 / 81.85 85.34 / 76.60 87.36 / 78.54 88.41 / 80.67 89.54 / 81.98 89.59 / 81.54 90.43 / 82.42 it → zhs 71.92 / 77.53 45.03 / 49.01 77.82 / 81.10 77.72 / 80.20 79.92 / 81.59 74.76 / 77.11 79.77 / 81.55 76.57 / 79.45 81.88 / 84.24 78.24 / 80.71 83.66 / 85.55 zhs → ja 84.50 / 88.80 78.86 / 84.43 88.28 / 90.49 88.26 / 90.83 88.48 / 90.77 86.00 / 88.47 88.15 / 89.66 88.53 / 90.35 89.61 / 90.98 89.20 / 90.81 89.96 / 91.07 ja → zhs 56.93 / 81.37 45.05 / 67.95 67.40 / 83.13 67.05 / 82.83 67.81 / 83.58 62.25 / 80.38 64.77 / 82.57 66.34 / 83.22 68.44 / 85.02 66.75 / 83.67 70.71 / 86.04 zhs → k 67.85 / 84.59 67.51 / 80.98 77.76 / 89.07 77.94 / 89.42 78.10 / 89.79 66.86 / 83.02 69.58 / 85.59 76.82 / 88.69 77.75 / 90.07 78.89 / 89.77 80.27 / 91.20 k → zhs 34.84 / 79.07 26.88 / 64.74 40.62 / 80.52 40.67 / 80.29 42.48 / 80.47 36.66 / 75.93 42.11 / 78.06 39.45 / 79.53 45.57 / 81.33 40.17 / 80.45 46.17 / 82.07 zhs → km 67.41 / 78.30 65.90 / 77.73 77.18 / 85.11 78.44 / 85.44 77.47 / 84.63 67.99 / 79.22 73.20 / 82.41 76.91 / 84.32 79.96 / 85.86 78.09 / 85.17 81.38 / 86.61 km → zhs 36.16 / 82.73 34.79 / 78.39 42.74 / 84.44 42.20 / 84.27 46.12 / 84.47 41.80 / 80.36 46.58 / 81.74 42.70 / 83.43 47.76 / 84.26 43.37 / 84.16 50.53 / 84.96 zhs → ko 91.22 / 88.95 87.75 / 85.58 93.47 / 90.83 93.88 / 91.23 94.23 / 91.18 88.33 / 87.45 91.06 / 89.01 93.49 / 90.63 94.54 / 91.48 94.29 / 91.19 95.30 / 91.90 ko → zhs 63.98 / 85.85 57.35 / 76.69 72.44 / 87.74 71.74 / 87.79 72.43 / 87.80 66.32 / 84.70 69.63 / 85.16 69.51 / 86.70 71.91 / 87.52 71.29 / 87.62 73.99 / 88.19 zhs → lo 65.22 / 72.21 73.30 / 76.72 79.05 / 82.11 80.44 / 82.61 77.36 / 80.93 68.24 / 74.95 73.04 / 78.58 79.34 / 81.33 82.54 / 83.47 80.56 / 82.30 84.67 / 84.46 lo → zhs 47.98 / 79.92 47.58 / 76.24 54.28 / 81.35 54.78 / 81.39 57.09 / 81.48 50.99 / 77.28 55.10 / 78.32 53.08 / 80.45 57.06 / 81.25 54.49 / 81.29 59.97 / 81.98 zhs → ms 84.99 / 75.63 85.74 / 76.52 90.67 / 81.70 90.56 / 81.62 91.45 / 82.59 88.06 / 79.44 90.69 / 82.48 90.82 / 81.73 93.14 / 84.90 91.14 / 82.42 94.05 / 85.51 ms → zhs 77.62 / 80.68 64.06 / 67.60 81.36 / 82.14 81.45 / 82.47 81.76 / 82.18 77.29 / 78.86 80.31 / 79.63 80.66 / 81.17 82.24 / 81.52 81.86 / 82.33 84.11 / 82.32 zhs → my 61.21 / 78.69 60.59 / 74.98 76.86 / 86.94 78.15 / 87.06 76.31 / 86.45 58.07 / 78.24 64.65 / 81.85 75.93 / 86.05 78.29 / 87.14 78.53 / 87.14 81.86 / 88.03 my → zhs 24.07 / 83.51 21.73 / 71.50 26.85 / 84.55 26.87 / 84.73 28.89 / 84.93 24.89 / 80.15 27.84 / 82.44 27.03 / 83.99 28.86 / 85.03 27.68 / 84.98 29.90 / 85.78 zhs → nb 91.99 / 85.50 88.73 / 81.50 94.28 / 87.38 94.11 / 87.68 94.54 / 88.24 89.35 / 81.67 91.13 / 83.87 93.82 / 86.50 94.62 / 88.09 94.23 / 87.58 95.77 / 89.23 nb → zhs 75.42 / 80.94 43.51 / 47.92 78.77 / 82.75 78.96 / 82.53 79.76 / 83.09 75.40 / 78.47 78.97 / 80.93 78.29 / 82.08 81.31 / 83.93 79.05 / 82.21 83.26 / 84.96 zhs → nl 90.98 / 83.16 87.67 / 79.26 92.82 / 84.59 93.02 / 84.64 93.48 / 84.97 88.87 / 78.57 90.68 / 81.09 92.68 / 83.44 93.84 / 85.10 93.21 / 84.76 94.63 / 86.69 nl → zhs 73.33 / 83.31 44.56 / 54.87 77.41 / 85.36 77.58 / 85.03 78.95 / 84.86 73.92 / 81.55 78.60 / 84.54 76.37 / 84.17 80.99 / 87.19 78.41 / 84.96 83.25 / 88.31 zhs → pl 86.73 / 79.18 83.12 / 75.68 89.77 / 81.50 90.18 / 81.82 90.76 / 82.43 82.53 / 74.29 84.72 / 77.65 88.54 / 80.34 89.24 / 82.63 90.11 / 81.50 91.23 / 83.62 pl → zhs 56.73 / 70.18 35.19 / 50.53 62.69 / 72.85 62.68 / 72.51 65.07 / 73.44 60.27 / 68.70 64.84 / 72.08 62.81 / 72.46 67.65 / 75.41 63.15 / 72.84 69.41 / 76.96 zhs → pt 89.21 / 78.65 87.32 / 75.97 91.11 / 79.90 91.59 / 80.22 91.71 / 80.39 88.93 / 77.06 91.07 / 78.87 91.57 / 79.64 92.38 / 81.21 92.15 / 80.56 93.36 / 81.71 pt → zhs 81.89 / 78.34 55.56 / 48.79 85.51 / 81.42 84.79 / 80.25 86.02 / 81.35 81.50 / 77.02 85.15 / 81.21 84.01 / 79.65 86.97 / 83.86 84.93 / 80.85 88.58 / 85.26 zhs → ro 83.39 / 83.02 80.93 / 80.06 87.02 / 85.58 87.16 / 85.91 87.99 / 86.66 80.61 / 79.62 82.23 / 81.86 85.75 / 84.97 86.38 / 86.57 86.69 / 86.15 87.81 / 87.22 ro → zhs 55.95 / 73.45 38.51 / 56.36 62.49 / 76.86 61.50 / 76.42 64.36 / 77.39 57.54 / 70.54 64.91 / 74.77 61.12 / 75.48 67.29 / 78.66 61.79 / 76.37 69.89 / 80.25 zhs → ru 91.50 / 77.75 90.12 / 76.52 93.31 / 79.51 93.33 / 80.11 93.00 / 79.84 87.98 / 74.20 90.37 / 76.11 92.53 / 79.05 93.06 / 79.35 93.20 / 79.94 93.89 / 79.93 ru → zhs 78.51 / 70.63 63.60 / 52.71 81.83 / 74.13 82.07 / 73.73 83.25 / 75.31 79.27 / 70.22 82.81 / 75.29 81.47 / 73.14 85.02 / 77.72 82.41 / 74.48 85.51 / 79.11 zhs → sk 83.29 / 84.45 78.58 / 79.97 87.47 / 86.92 87.30 / 87.25 88.06 / 87.89 77.88 / 78.49 82.07 / 82.86 86.23 / 86.06 87.07 / 88.12 87.40 / 87.44 88.01 / 88.69 sk → zhs 56.13 / 81.44 35.25 / 59.36 61.87 / 83.49 61.02 / 82.91 63.23 / 83.36 56.86 / 78.08 62.53 / 80.61 61.03 / 82.10 64.96 / 83.74 62.36 / 83.01 67.96 / 84.94 zhs → sl 81.52 / 82.91 77.74 / 78.94 85.15 / 85.11 86.12 / 85.53 86.92 / 86.18 77.24 / 76.17 79.94 / 80.16 83.72 / 83.16 85.51 / 85.54 85.56 / 84.98 86.84 / 87.20 sl → zhs 56.88 / 79.45 37.56 / 56.58 62.60 / 82.05 62.46 / 81.45 64.61 / 81.93 59.15 / 76.59 63.48 / 79.47 62.27 / 80.65 66.76 / 82.69 64.05 / 81.80 68.54 / 83.52 zhs → sv 92.51 / 85.32 88.72 / 80.50 94.22 / 87.41 94.90 / 87.70 94.52 / 87.86 89.35 / 80.74 91.73 / 83.43 93.83 / 86.13 94.74 / 87.93 94.64 / 87.56 95.58 / 89.02 sv → zhs 77.39 / 83.39 50.50 / 59.45 80.41 / 85.57 80.74 / 85.66 81.52 / 85.61 75.79 / 80.99 79.69 / 83.78 79.76 / 84.37 83.43 / 86.35 81.36 / 85.63 84.70 / 87.58 zhs → ta 76.71 / 69.98 74.74 / 64.46 83.60 / 76.18 84.73 / 77.34 82.83 / 76.92 69.06 / 67.86 75.58 / 72.11 80.54 / 75.07 84.09 / 78.11 82.83 / 76.46 86.67 / 79.29 ta → zhs 47.02 / 70.10 39.74 / 63.53 51.73 / 72.49 51.91 / 72.74 53.95 / 72.55 46.56 / 67.10 51.54 / 68.00 51.30 / 71.47 56.84 / 71.19 52.71 / 71.95 58.80 / 72.58 zhs → th 80.12 / 78.46 74.62 / 72.25 86.93 / 83.74 86.84 / 83.51 87.02 / 83.48 79.23 / 77.89 83.76 / 81.98 85.88 / 82.75 89.08 / 85.54 86.88 / 83.61 90.34 / 86.58 th → zhs 73.17 / 81.58 59.14 / 69.76 77.64 / 83.37 78.12 / 83.46 78.56 / 83.36 73.65 / 79.98 76.43 / 80.13 77.71 / 82.74 79.03 / 82.51 78.71 / 83.34 80.10 / 83.12 zhs → tl 79.63 / 77.98 78.77 / 75.91 85.86 / 82.32 86.55 / 82.42 85.40 / 82.43 78.15 / 77.55 83.24 / 80.89 84.81 / 81.71 88.32 / 83.96 85.58 / 82.56 89.36 / 84.96 tl → zhs 46.86 / 79.32 38.67 / 65.53 50.56 / 81.12 51.19 / 81.26 52.62 / 81.20 48.04 / 77.03 51.55 / 78.14 50.42 / 79.92 52.69 / 81.06 51.60 / 81.23 54.44 / 81.67 zhs → tr 79.91 / 82.98 76.35 / 77.77 85.22 / 85.22 85.28 / 85.57 86.25 / 85.94 77.09 / 78.67 81.20 / 82.29 83.49 / 83.91 86.46 / 86.17 85.48 / 85.36 87.83 / 87.46 tr → zhs 56.91 / 80.69 37.38 / 63.44 64.91 / 82.29 64.81 / 82.26 68.02 / 82.39 57.79 / 77.01 63.94 / 79.17 63.12 / 80.94 68.16 / 82.26 64.57 / 82.68 70.23 / 83.47 zhs → ur 77.02 / 69.67 78.38 / 70.42 87.56 / 77.37 87.89 / 78.27 88.85 / 79.20 74.63 / 69.87 80.51 / 74.11 84.65 / 76.16 88.06 / 79.20 87.18 / 78.10 90.45 / 80.75 ur → zhs 53.17 / 72.69 46.37 / 67.76 59.76 / 73.97 59.03 / 74.38 60.37 / 73.85 52.83 / 69.52 58.26 / 70.36 58.00 / 72.94 62.52 / 73.03 59.43 / 74.35 65.05 / 74.25 zhs → uz 63.97 / 80.61 61.35 / 75.45 74.69 / 86.12 75.03 / 86.41 74.80 / 85.99 64.05 / 79.33 69.58 / 83.56 72.66 / 84.86 76.53 / 87.57 75.35 / 86.08 78.73 / 88.51 uz → zhs 26.59 / 80.35 23.62 / 73.52 32.05 / 81.83 31.70 / 82.06 34.00 / 82.29 29.11 / 76.89 32.55 / 78.16 31.96 / 80.84 34.99 / 81.84 32.07 / 82.14 35.90 / 82.69 zhs → vi 90.28 / 83.49 87.67 / 80.41 91.89 / 85.73 92.37 / 86.01 92.81 / 86.24 89.28 / 83.25 91.39 / 86.00 92.09 / 85.55 93.78 / 87.77 92.63 / 86.47 94.60 / 88.53 vi → zhs 80.54 / 83.75 66.91 / 69.09 85.08 / 85.55 84.85 / 85.13 85.76 / 85.43 80.97 / 82.05 84.02 / 82.64 84.16 / 84.76 86.18 / 84.89 85.12 / 85.33 87.35 / 85.40 zhs → yue 94.05 / 60.77 83.85 / 64.46 93.28 / 67.37 92.07 / 69.14 94.29 / 70.13 93.88 / 63.45 93.62 / 76.32 94.96 / 63.79 95.48 / 76.40 95.29 / 64.37 95.67 / 78.51 yue → zhs 94.05 / 63.92 88.64 / 61.56 95.09 / 67.01 95.23 / 67.99 95.54 / 72.66 93.98 / 76.09 94.90 / 80.69 95.49 / 77.24 96.50 / 81.86 95.42 / 78.12 96.64 / 82.44 zhs → zht 96.07 / 27.19 70.59 / 42.73 96.53 / 22.24 96.46 / 21.51 96.42 / 22.53 95.92 / 45.44 96.26 / 59.65 96.75 / 40.81 97.42 / 63.50 96.86 / 42.93 97.79 / 66.67 zht → zhs 92.59 / 22.11 83.38 / 69.54 92.87 / 23.97 92.83 / 24.27 93.04 / 25.91 93.91 / 56.99 94.71 / 69.56 94.63 / 54.94 96.15 / 70.18 94.84 / 55.33 94.92 / 69.91 Table 11: Reference-free Chinese-centric evaluation results (XCOMET / COMETKiwi) of baseline models (Part I) and MiLMMT models on the FLORES+ benchmark. D.2 Reference-Based Results Tables 12–15 report the per-direction reference-based FLORES+ results (spBLEU / XCOMET). Direction TranslateGemma-4B TranslateGemma-12B TranslateGemma-27B GemmaX2-28-2B GemmaX2-28-9B Hunyuan-MT-7B HY-MT1.5-1.8B HY-MT1.5-7B HY-MT2-1.8B HY-MT2-7B HY-MT2-30B-A3B Seed-X-Instruct-7B Seed-X-PPO-7B Tower-Plus-2B Tower-Plus-9B Tower-Plus-72B en → ar 31.42 / 92.06 33.73 / 94.89 33.79 / 96.20 37.71 / 90.33 41.91 / 93.56 32.27 / 95.46 27.96 / 92.30 32.68 / 95.57 34.00 / 92.05 38.39 / 94.93 38.80 / 95.37 41.78 / 92.87 42.62 / 93.38 - / - - / - - / - ar → en 34.36 / 89.18 36.37 / 92.60 40.06 / 94.09 45.26 / 92.01 48.05 / 94.18 32.52 / 92.22 26.51 / 89.55 33.42 / 92.72 34.33 / 89.79 42.98 / 93.51 44.06 / 94.45 47.05 / 92.38 43.72 / 94.36 - / - - / - - / - en → az 6.82 / 39.69 18.27 / 78.24 21.93 / 86.27 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - az → en 20.84 / 76.12 23.68 / 85.74 26.14 / 88.98 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → bn 24.35 / 87.37 24.81 / 91.23 26.10 / 92.14 31.60 / 88.23 33.13 / 90.80 24.89 / 92.34 22.67 / 88.91 25.77 / 92.72 25.82 / 88.82 28.90 / 92.54 27.63 / 93.28 - / - - / - - / - - / - - / - bn → en 28.66 / 85.75 31.33 / 89.90 34.53 / 91.73 39.77 / 89.28 42.80 / 91.36 25.16 / 85.21 21.36 / 85.78 28.52 / 86.98 27.55 / 85.04 35.89 / 89.31 37.36 / 91.97 - / - - / - - / - - / - - / - en → bg 37.72 / 92.11 39.33 / 96.08 40.82 / 97.21 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - bg → en 37.76 / 92.29 39.51 / 94.25 42.45 / 95.71 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → ca 38.42 / 91.34 42.09 / 95.57 39.11 / 96.30 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - ca → en 42.43 / 92.31 44.23 / 94.61 46.79 / 95.66 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → cs 31.32 / 89.73 34.86 / 94.60 34.17 / 95.36 40.65 / 91.87 43.74 / 94.23 38.43 / 94.77 33.69 / 90.20 38.16 / 94.70 38.83 / 90.52 43.00 / 94.87 42.99 / 95.54 40.76 / 93.64 44.51 / 94.48 39.95 / 91.66 43.28 / 94.33 42.19 / 93.64 cs → en 37.57 / 93.31 38.32 / 95.06 41.36 / 95.92 46.16 / 95.28 47.78 / 96.08 35.53 / 94.64 30.46 / 93.45 37.11 / 94.52 38.77 / 93.47 44.18 / 95.75 44.35 / 96.26 44.89 / 93.35 45.32 / 96.13 44.70 / 94.73 47.56 / 95.90 47.41 / 96.17 en → zhs 31.02 / 91.99 31.96 / 94.45 33.60 / 95.15 37.94 / 92.88 39.85 / 93.72 31.03 / 96.04 30.57 / 93.81 29.94 / 95.58 35.79 / 93.47 36.01 / 95.64 37.68 / 95.71 37.60 / 94.01 34.09 / 95.28 36.39 / 90.84 39.92 / 93.32 42.21 / 94.51 zhs → en 27.24 / 95.63 28.90 / 96.71 30.94 / 97.27 34.68 / 96.48 36.57 / 96.96 29.57 / 96.86 23.89 / 96.00 27.48 / 96.82 26.95 / 96.17 28.98 / 97.04 29.82 / 97.68 36.47 / 96.46 33.19 / 96.86 32.02 / 95.53 34.78 / 96.77 36.72 / 97.26 en → zht 12.41 / 89.41 12.04 / 92.73 27.31 / 94.66 - / - - / - 8.43 / 94.26 8.44 / 92.63 24.16 / 94.27 10.72 / 91.19 29.46 / 94.64 29.71 / 95.18 - / - - / - 30.11 / 88.95 33.01 / 92.11 31.59 / 93.14 zht → en 25.80 / 93.78 27.46 / 95.70 29.78 / 96.49 - / - - / - 27.86 / 95.48 22.01 / 93.96 27.98 / 95.33 27.07 / 93.67 32.18 / 95.77 33.27 / 96.62 - / - - / - 29.92 / 93.99 33.39 / 96.06 35.84 / 96.62 en → da 42.35 / 95.94 43.94 / 97.72 41.86 / 98.22 - / - - / - - / - - / - - / - - / - - / - - / - 51.01 / 97.19 54.91 / 97.38 50.91 / 96.01 53.66 / 97.16 51.88 / 96.65 da → en 45.10 / 95.11 46.14 / 96.18 47.73 / 97.13 - / - - / - - / - - / - - / - - / - - / - - / - 54.25 / 95.61 53.33 / 97.22 53.28 / 96.27 55.60 / 97.26 54.95 / 97.25 en → de 39.21 / 97.73 40.28 / 98.50 41.85 / 98.89 45.58 / 97.66 49.34 / 98.26 40.68 / 98.60 35.01 / 97.66 41.18 / 98.58 40.89 / 97.48 46.03 / 98.42 46.78 / 98.72 49.02 / 98.30 49.72 / 97.95 44.95 / 97.50 48.62 / 98.38 48.11 / 98.43 de → en 41.12 / 95.49 42.10 / 96.49 44.22 / 97.05 49.53 / 96.72 50.85 / 97.12 39.01 / 96.47 33.40 / 95.93 40.03 / 96.30 42.34 / 95.61 47.24 / 97.01 47.59 / 97.39 50.92 / 96.28 49.19 / 97.13 48.38 / 96.22 50.24 / 97.18 51.11 / 97.33 en → el 30.38 / 88.80 34.32 / 93.37 35.38 / 95.49 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - el → en 33.40 / 90.76 35.27 / 93.40 37.86 / 94.73 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → tl 30.08 / 86.91 32.59 / 90.38 35.55 / 90.46 38.05 / 87.46 40.11 / 89.36 26.18 / 84.06 25.67 / 87.60 26.74 / 84.10 32.79 / 87.79 36.27 / 90.84 35.48 / 91.17 - / - - / - - / - - / - - / - tl → en 39.74 / 91.04 40.59 / 93.52 45.31 / 94.87 52.76 / 93.48 55.75 / 94.68 35.20 / 92.49 30.05 / 91.40 37.00 / 92.61 40.61 / 90.82 48.18 / 93.76 48.79 / 94.59 - / - - / - - / - - / - - / - en → fi 27.19 / 87.98 30.98 / 94.05 32.21 / 95.63 - / - - / - - / - - / - - / - - / - - / - - / - 43.49 / 93.72 45.60 / 94.74 32.93 / 89.80 37.95 / 94.42 35.31 / 92.43 fi → en 33.37 / 90.81 35.72 / 94.06 37.14 / 95.34 - / - - / - - / - - / - - / - - / - - / - - / - 41.75 / 93.24 39.90 / 95.10 38.80 / 92.84 42.19 / 95.34 42.34 / 95.55 en → fr 47.04 / 93.78 47.47 / 95.77 44.06 / 96.17 54.59 / 93.97 57.60 / 95.74 44.54 / 95.53 39.72 / 93.16 44.60 / 95.84 47.94 / 93.34 52.04 / 95.77 51.75 / 96.07 57.41 / 95.19 57.91 / 95.15 54.32 / 93.61 56.14 / 95.79 56.85 / 95.74 fr → en 42.48 / 94.60 42.79 / 95.35 45.11 / 96.11 50.27 / 96.18 51.34 / 96.54 38.67 / 95.53 33.00 / 94.62 39.79 / 95.42 43.00 / 95.27 47.24 / 96.28 48.15 / 96.54 49.71 / 93.94 49.71 / 96.62 49.41 / 96.00 50.49 / 96.47 51.76 / 96.52 en → he 29.39 / 84.93 34.42 / 93.30 37.17 / 95.24 42.71 / 89.97 47.20 / 93.46 29.60 / 88.09 25.78 / 84.36 30.48 / 88.73 34.87 / 86.64 42.37 / 93.34 42.47 / 94.57 - / - - / - - / - - / - - / - he → en 36.88 / 90.28 40.06 / 93.98 43.73 / 95.79 48.36 / 94.32 50.92 / 95.79 34.08 / 91.60 27.92 / 89.12 35.90 / 92.11 36.36 / 89.04 45.64 / 94.56 46.72 / 95.64 - / - - / - - / - - / - - / - en → hi 29.72 / 85.62 29.23 / 90.21 31.76 / 91.65 36.68 / 83.39 38.79 / 87.23 24.54 / 91.84 22.88 / 88.97 26.20 / 92.10 28.21 / 87.16 30.88 / 91.78 30.48 / 92.35 - / - - / - 36.13 / 79.98 40.39 / 84.41 38.12 / 83.71 hi → en 33.88 / 89.75 35.39 / 93.09 39.92 / 94.15 45.48 / 92.51 48.01 / 94.00 30.82 / 91.49 25.79 / 90.26 33.16 / 92.31 33.50 / 89.85 41.97 / 93.29 42.81 / 94.37 - / - - / - 42.85 / 91.34 47.53 / 93.70 47.30 / 93.98 en → hr 25.59 / 89.38 31.64 / 95.58 33.72 / 96.90 - / - - / - - / - - / - - / - - / - - / - - / - 38.46 / 92.86 41.51 / 94.23 - / - - / - - / - hr → en 35.74 / 92.11 38.51 / 94.66 39.80 / 95.85 - / - - / - - / - - / - - / - - / - - / - - / - 41.69 / 90.63 44.76 / 95.48 - / - - / - - / - en → hu 24.07 / 84.46 28.05 / 93.23 30.57 / 96.15 - / - - / - - / - - / - - / - - / - - / - - / - 36.17 / 94.97 39.40 / 95.97 34.29 / 91.76 37.58 / 95.61 35.92 / 94.30 hu → en 32.14 / 90.27 35.52 / 93.63 37.26 / 94.88 - / - - / - - / - - / - - / - - / - - / - - / - 42.45 / 92.70 41.52 / 95.34 40.10 / 93.30 44.11 / 94.98 43.87 / 95.37 en → id 38.78 / 96.07 38.26 / 97.23 39.09 / 97.68 49.39 / 95.71 50.91 / 96.27 32.86 / 97.48 26.80 / 96.02 33.30 / 97.46 37.11 / 95.60 41.56 / 97.23 41.52 / 97.34 50.50 / 95.46 51.55 / 95.42 - / - - / - - / - id → en 40.22 / 94.54 41.05 / 95.57 44.22 / 96.22 50.36 / 95.49 51.98 / 96.12 36.63 / 94.78 30.49 / 93.57 37.55 / 94.88 40.16 / 93.66 45.68 / 95.77 46.72 / 96.10 49.91 / 94.82 48.26 / 96.00 - / - - / - - / - en → it 34.39 / 95.47 36.88 / 96.58 33.00 / 97.02 37.83 / 95.36 40.07 / 96.15 31.18 / 96.98 27.23 / 94.79 32.16 / 96.91 32.38 / 94.56 35.45 / 96.38 36.95 / 97.11 38.90 / 95.48 38.72 / 95.15 37.32 / 94.55 39.85 / 96.35 39.92 / 96.59 it → en 34.70 / 95.02 34.83 / 95.76 36.58 / 96.34 39.34 / 95.90 40.13 / 96.71 32.36 / 95.45 27.17 / 94.59 32.84 / 95.39 34.56 / 94.70 38.28 / 96.06 38.94 / 96.66 41.26 / 92.85 39.38 / 96.58 38.77 / 95.62 40.90 / 96.29 41.48 / 96.62 en → ja 26.16 / 89.40 26.60 / 92.63 27.54 / 93.55 31.31 / 91.64 34.06 / 93.91 29.58 / 94.91 25.35 / 93.38 29.74 / 94.79 29.65 / 92.58 32.36 / 94.83 33.00 / 95.41 37.48 / 92.29 39.17 / 93.84 30.97 / 90.97 33.86 / 93.82 35.10 / 94.24 ja → en 25.88 / 91.26 27.53 / 94.42 29.61 / 95.74 33.45 / 93.95 35.48 / 95.53 26.57 / 94.58 21.18 / 92.95 26.99 / 94.83 25.67 / 93.12 31.45 / 95.33 32.39 / 95.77 34.16 / 92.23 31.93 / 95.31 31.37 / 92.92 34.57 / 95.28 35.82 / 95.58 en → k 10.31 / 42.15 22.40 / 74.71 27.21 / 84.43 - / - - / - 0.51 / 32.27 1.55 / 44.45 0.25 / 32.26 0.96 / 29.27 2.74 / 43.87 0.57 / 34.34 - / - - / - - / - - / - - / - k → en 25.71 / 65.46 29.83 / 77.19 34.40 / 81.97 - / - - / - 15.62 / 47.41 10.87 / 32.71 19.13 / 50.59 14.43 / 34.52 28.04 / 65.16 36.20 / 82.26 - / - - / - - / - - / - - / - en → km 3.01 / 29.13 14.80 / 64.33 19.72 / 74.41 24.13 / 76.19 26.64 / 80.93 17.84 / 72.03 19.53 / 74.37 18.21 / 71.79 24.04 / 79.63 25.87 / 83.56 26.19 / 83.52 - / - - / - - / - - / - - / - km → en 23.91 / 63.20 28.37 / 73.44 32.35 / 77.06 36.65 / 77.85 39.68 / 81.51 21.87 / 71.19 20.75 / 75.15 25.46 / 72.28 27.54 / 76.11 34.72 / 79.99 37.06 / 80.63 - / - - / - - / - - / - - / - en → ko 22.73 / 91.14 23.85 / 95.09 25.70 / 95.85 26.60 / 91.56 29.83 / 94.21 24.57 / 95.85 21.53 / 94.04 25.71 / 95.78 26.34 / 93.19 27.92 / 95.49 28.77 / 95.40 25.89 / 88.97 28.04 / 90.90 26.90 / 90.38 29.92 / 94.28 30.05 / 94.00 ko → en 26.95 / 90.53 29.52 / 93.63 31.92 / 94.89 35.99 / 93.02 38.44 / 94.83 28.78 / 94.07 23.62 / 92.52 29.58 / 94.21 28.99 / 92.37 34.57 / 94.60 35.64 / 95.17 35.81 / 91.53 34.47 / 93.84 33.46 / 91.86 37.54 / 94.64 38.88 / 95.00 en → lo 5.07 / 32.86 20.23 / 72.58 26.28 / 81.79 29.27 / 79.40 33.53 / 84.63 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - lo → en 26.42 / 70.05 31.22 / 81.64 36.21 / 84.99 40.14 / 83.90 44.11 / 87.43 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → my 4.86 / 34.08 11.11 / 56.01 15.78 / 70.05 20.96 / 70.97 25.93 / 80.14 17.37 / 68.68 19.30 / 70.72 18.67 / 70.09 24.17 / 76.39 26.64 / 82.80 26.44 / 83.02 - / - - / - - / - - / - - / - my → en 17.34 / 57.34 22.82 / 71.01 27.14 / 76.71 30.69 / 74.97 36.12 / 80.96 17.83 / 66.23 18.02 / 72.72 21.43 / 69.24 25.47 / 73.57 30.47 / 79.41 32.51 / 79.93 - / - - / - - / - - / - - / - en → nl 29.71 / 95.60 31.24 / 97.19 31.37 / 97.83 35.53 / 96.05 37.17 / 97.08 27.80 / 96.12 26.09 / 94.72 28.20 / 96.14 31.38 / 94.81 34.49 / 97.01 34.91 / 97.66 42.49 / 97.29 43.50 / 97.40 35.22 / 95.66 37.64 / 97.50 36.72 / 97.35 nl → en 33.30 / 94.80 34.03 / 95.86 35.25 / 96.53 37.87 / 95.81 38.79 / 96.38 30.20 / 95.14 26.57 / 94.33 31.20 / 95.16 32.28 / 94.67 36.98 / 96.12 37.22 / 96.60 37.18 / 92.03 39.24 / 96.75 37.54 / 95.54 39.76 / 96.41 39.39 / 96.51 en → nb 45.40 / 96.17 47.10 / 97.60 42.70 / 98.05 - / - - / - - / - - / - - / - - / - - / - - / - 45.60 / 94.71 49.80 / 95.93 65.60 / 95.94 70.61 / 96.11 69.10 / 95.58 nb → en 53.30 / 94.64 53.77 / 95.67 57.73 / 96.79 - / - - / - - / - - / - - / - - / - - / - - / - 62.85 / 94.69 66.03 / 96.41 65.83 / 95.80 70.62 / 97.01 68.68 / 96.87 en → fa 24.36 / 88.40 25.95 / 93.97 27.54 / 95.24 34.26 / 89.77 36.68 / 93.27 26.22 / 92.04 23.18 / 89.05 26.93 / 92.34 28.81 / 89.01 33.12 / 93.55 32.14 / 94.21 - / - - / - - / - - / - - / - fa → en 32.07 / 88.34 33.42 / 91.96 37.04 / 93.93 42.34 / 92.35 44.14 / 93.66 29.68 / 90.45 23.79 / 87.51 31.62 / 90.96 30.63 / 87.53 38.54 / 92.24 40.46 / 93.63 - / - - / - - / - - / - - / - en → pl 26.84 / 92.15 28.89 / 95.53 29.82 / 96.53 32.59 / 92.67 34.81 / 95.27 25.48 / 92.28 25.90 / 89.49 25.92 / 93.15 30.35 / 90.47 33.27 / 95.22 33.96 / 96.62 38.52 / 95.24 40.62 / 95.97 31.51 / 92.85 34.87 / 95.79 33.93 / 95.50 pl → en 31.04 / 93.50 32.32 / 94.83 33.26 / 95.99 36.26 / 94.96 37.06 / 95.96 28.88 / 93.67 24.67 / 92.75 30.18 / 93.93 30.45 / 92.61 35.46 / 95.27 35.55 / 95.91 36.18 / 90.34 36.65 / 95.90 35.46 / 94.31 37.16 / 96.02 38.24 / 96.01 en → pt 45.09 / 96.76 45.26 / 97.42 39.76 / 97.57 53.65 / 96.57 54.84 / 96.98 45.08 / 97.53 40.22 / 96.52 46.01 / 97.47 50.86 / 96.41 54.06 / 97.21 52.16 / 97.60 57.60 / 96.64 51.28 / 96.26 50.24 / 95.72 49.21 / 96.80 50.13 / 97.01 pt → en 44.23 / 95.08 45.27 / 95.74 48.24 / 96.43 55.02 / 96.41 56.62 / 96.59 40.91 / 95.34 35.07 / 94.95 41.84 / 95.68 45.87 / 94.84 51.34 / 96.34 51.52 / 96.80 56.06 / 95.65 54.15 / 96.63 53.43 / 95.90 55.80 / 96.64 57.02 / 96.91 en → ro 38.74 / 92.41 40.76 / 95.50 42.09 / 96.65 - / - - / - - / - - / - - / - - / - - / - - / - 53.90 / 95.29 56.98 / 95.39 47.14 / 92.83 49.97 / 95.80 48.40 / 94.79 ro → en 41.39 / 93.18 41.64 / 94.31 44.27 / 95.85 - / - - / - - / - - / - - / - - / - - / - - / - 47.61 / 90.42 49.80 / 95.95 47.58 / 94.37 49.91 / 95.90 50.70 / 96.16 en → ru 33.29 / 93.77 33.91 / 96.09 32.50 / 96.88 38.20 / 92.08 41.75 / 94.72 31.92 / 95.62 29.66 / 92.89 32.15 / 95.52 35.76 / 91.38 39.44 / 95.20 40.33 / 96.22 42.88 / 95.17 44.25 / 95.27 39.12 / 92.31 41.81 / 95.26 41.90 / 95.42 ru → en 33.83 / 93.62 35.17 / 95.17 37.43 / 96.23 40.76 / 95.17 42.94 / 96.16 32.20 / 95.33 27.07 / 93.95 33.76 / 95.46 34.00 / 93.76 39.78 / 95.59 40.57 / 96.40 42.08 / 93.53 40.36 / 96.25 39.79 / 94.44 41.94 / 95.98 43.26 / 96.32 en → sk 28.77 / 86.70 33.24 / 94.22 33.73 / 95.37 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - sk → en 36.35 / 92.08 38.18 / 94.56 41.10 / 95.58 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → sl 23.64 / 81.75 29.35 / 92.92 31.37 / 95.34 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - sl → en 33.65 / 90.16 36.03 / 93.86 37.50 / 95.28 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → es 32.06 / 96.63 33.76 / 97.22 33.44 / 97.40 33.47 / 96.01 34.87 / 96.53 32.48 / 97.27 30.03 / 96.56 33.19 / 97.29 32.92 / 96.10 34.74 / 96.92 35.17 / 97.57 34.72 / 96.22 34.66 / 96.27 33.65 / 95.56 35.28 / 97.08 35.69 / 97.13 es → en 32.63 / 94.60 33.29 / 95.86 34.39 / 96.42 36.13 / 95.71 37.29 / 96.26 30.42 / 95.46 25.84 / 94.70 31.24 / 95.66 32.04 / 94.47 36.13 / 95.87 37.03 / 96.27 39.26 / 92.55 36.13 / 96.37 35.94 / 95.39 37.83 / 96.09 38.52 / 96.25 en → sv 42.33 / 95.87 42.55 / 97.97 41.01 / 98.30 - / - - / - - / - - / - - / - - / - - / - - / - 56.31 / 97.30 58.95 / 97.37 50.53 / 96.10 52.99 / 97.57 51.76 / 96.99 sv → en 44.19 / 94.71 44.86 / 95.80 47.53 / 96.59 - / - - / - - / - - / - - / - - / - - / - - / - 50.94 / 93.76 54.60 / 96.73 52.21 / 95.81 54.52 / 96.73 54.46 / 96.84 en → ta 27.41 / 80.65 25.91 / 85.82 27.59 / 86.91 - / - - / - 19.81 / 77.66 21.81 / 80.20 21.12 / 79.57 25.66 / 79.03 28.16 / 85.69 28.30 / 86.98 - / - - / - - / - - / - - / - ta → en 26.01 / 81.23 29.09 / 86.58 33.29 / 89.66 - / - - / - 22.62 / 79.46 18.96 / 78.97 25.63 / 81.87 24.32 / 79.83 33.43 / 88.13 35.05 / 89.57 - / - - / - - / - - / - - / - en → th 34.95 / 90.88 35.42 / 94.98 38.20 / 96.05 40.85 / 91.60 44.18 / 94.34 35.31 / 95.52 32.01 / 92.03 36.75 / 95.50 38.23 / 92.05 41.68 / 95.44 41.44 / 95.72 43.30 / 91.60 45.47 / 92.26 - / - - / - - / - th → en 28.32 / 91.54 29.84 / 94.53 32.93 / 95.57 35.68 / 93.77 40.01 / 95.52 26.56 / 94.23 21.10 / 92.37 28.24 / 94.53 27.46 / 91.95 34.76 / 95.09 36.76 / 95.79 37.61 / 90.80 35.59 / 95.14 - / - - / - - / - en → tr 28.00 / 88.55 30.05 / 93.14 33.67 / 94.54 38.96 / 90.31 42.17 / 93.20 32.49 / 94.32 26.94 / 91.93 32.52 / 94.74 32.55 / 91.24 36.95 / 94.00 36.71 / 95.19 44.54 / 90.78 46.73 / 91.93 - / - - / - - / - tr → en 34.01 / 89.34 36.48 / 92.81 38.99 / 94.54 45.22 / 93.43 47.27 / 94.71 32.65 / 92.07 28.52 / 91.03 34.02 / 92.12 35.51 / 91.19 42.03 / 94.34 42.91 / 94.86 44.25 / 91.43 44.37 / 94.52 - / - - / - - / - en → ur 16.66 / 71.80 21.01 / 86.86 24.33 / 90.81 28.90 / 84.25 31.19 / 88.28 18.54 / 85.98 19.62 / 84.25 19.78 / 86.13 23.73 / 84.55 27.15 / 90.49 26.94 / 91.43 - / - - / - - / - - / - - / - ur → en 27.02 / 81.41 31.09 / 87.99 34.77 / 91.32 39.68 / 89.39 43.23 / 91.80 25.70 / 85.89 20.64 / 81.11 28.64 / 86.66 26.49 / 80.47 36.46 / 89.58 38.47 / 91.44 - / - - / - - / - - / - - / - en → uz 0.28 / 21.57 21.16 / 74.02 21.02 / 82.52 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - uz → en 25.52 / 66.64 30.53 / 77.44 35.17 / 82.97 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - en → vi 36.07 / 93.24 35.17 / 95.18 36.48 / 96.08 44.14 / 93.09 45.78 / 94.21 32.89 / 95.90 28.66 / 94.43 33.23 / 95.94 37.34 / 93.14 38.38 / 95.57 39.31 / 95.94 47.43 / 92.54 48.67 / 93.07 - / - - / - - / - vi → en 32.34 / 93.16 34.45 / 94.92 36.75 / 95.74 43.05 / 94.68 44.48 / 95.77 31.86 / 94.91 25.91 / 93.66 32.73 / 94.91 33.89 / 93.46 39.99 / 95.47 40.94 / 95.82 40.69 / 92.49 42.43 / 95.79 - / - - / - - / - en → yue 18.24 / 89.45 20.18 / 93.41 21.03 / 94.18 - / - - / - 21.19 / 93.13 1.72 / 80.04 23.91 / 93.00 22.14 / 88.97 29.06 / 93.52 31.53 / 93.06 - / - - / - - / - - / - - / - yue → en 26.41 / 92.43 28.21 / 94.85 31.38 / 95.66 - / - - / - 29.69 / 95.08 23.15 / 92.77 30.74 / 95.13 29.11 / 92.66 35.45 / 95.17 36.26 / 95.87 - / - - / - - / - - / - - / - en → ms 31.15 / 92.58 32.31 / 95.84 36.83 / 96.41 43.71 / 94.73 44.97 / 95.43 31.38 / 96.41 24.82 / 95.76 31.42 / 96.70 34.73 / 95.22 38.89 / 96.30 38.02 / 96.79 48.25 / 93.44 48.39 / 94.57 - / - - / - - / - ms → en 38.83 / 93.10 40.61 / 94.84 44.52 / 95.66 51.02 / 95.15 52.34 / 95.91 36.59 / 94.15 29.80 / 92.90 37.31 / 94.56 39.25 / 92.73 45.63 / 95.35 47.01 / 95.89 48.51 / 93.82 49.01 / 95.60 - / - - / - - / - Table 12: Reference-based English-centric evaluation results (spBLEU / XCOMET) of baseline models (Part I) on the FLORES+ benchmark. Direction Google Translate NLLB-54.5B Gemini 2.5 Pro Gemini 3 Pro GPT-5 MiLMMT-46-1B-v0.1 MiLMMT-46-1B-v1.0 MiLMMT-46-4B-v0.1 MiLMMT-46-4B-v1.0 MiLMMT-46-12B-v0.1 MiLMMT-46-12B-v1.0 en → ar 44.90 / 95.14 43.03 / 90.90 41.55 / 95.43 43.60 / 95.78 38.68 / 94.69 34.99 / 87.04 33.66 / 88.67 40.31 / 93.00 37.48 / 93.95 42.61 / 94.34 38.48 / 95.90 ar → en 50.99 / 94.87 48.26 / 92.50 49.55 / 95.54 49.26 / 95.67 46.74 / 95.10 42.43 / 90.13 41.25 / 90.60 46.94 / 93.51 45.54 / 93.63 48.92 / 94.78 47.23 / 94.96 en → az 26.51 / 82.61 24.63 / 79.16 27.41 / 90.16 28.07 / 90.47 25.85 / 89.46 22.27 / 76.34 21.73 / 79.44 25.64 / 86.38 24.96 / 88.19 26.95 / 88.76 25.89 / 90.61 az → en 33.24 / 90.76 31.94 / 87.01 31.77 / 91.71 32.12 / 91.82 29.73 / 91.43 26.22 / 84.35 25.31 / 86.04 29.18 / 89.82 28.23 / 90.22 30.49 / 90.89 29.24 / 91.71 en → bn 37.55 / 88.41 36.04 / 85.21 33.47 / 90.64 32.86 / 92.28 28.46 / 89.72 27.62 / 83.29 25.69 / 86.41 31.66 / 89.69 28.59 / 92.01 32.73 / 91.35 29.01 / 93.51 bn → en 44.97 / 91.90 42.22 / 89.21 44.35 / 92.95 44.18 / 93.28 41.50 / 93.03 36.85 / 86.34 36.20 / 87.79 41.63 / 90.72 40.39 / 91.37 43.04 / 91.92 41.73 / 92.52 en → bg 53.44 / 95.44 49.98 / 92.10 51.39 / 96.37 52.29 / 96.67 47.61 / 95.70 45.61 / 90.51 43.85 / 91.80 50.79 / 95.03 47.97 / 95.89 51.87 / 95.99 47.29 / 96.79 bg → en 51.32 / 95.77 47.01 / 93.54 49.37 / 95.97 50.52 / 96.21 47.86 / 95.65 45.53 / 93.52 44.19 / 94.01 48.59 / 95.45 47.04 / 95.80 49.75 / 95.87 47.88 / 96.05 en → ca 52.33 / 94.98 48.85 / 92.54 50.96 / 95.73 52.19 / 95.50 46.53 / 95.40 46.20 / 92.63 44.31 / 93.51 49.40 / 94.87 45.58 / 95.51 50.78 / 95.59 47.46 / 96.47 ca → en 55.13 / 95.67 52.58 / 93.77 53.02 / 96.02 54.53 / 96.33 52.30 / 96.08 49.90 / 94.05 48.86 / 94.42 52.34 / 95.61 51.10 / 95.89 53.84 / 96.18 52.40 / 96.27 en → cs 46.66 / 94.30 42.38 / 90.10 44.94 / 95.55 46.44 / 95.62 43.66 / 95.67 37.56 / 87.71 36.82 / 89.82 42.69 / 93.73 40.96 / 94.92 44.91 / 95.14 41.89 / 95.79 cs → en 49.97 / 96.15 45.31 / 94.15 47.78 / 96.35 48.71 / 96.56 46.31 / 96.11 44.22 / 94.02 42.85 / 94.59 46.94 / 95.94 45.17 / 96.22 48.39 / 96.37 46.67 / 96.58 en → zhs 44.05 / 93.84 26.61 / 76.72 39.42 / 94.26 40.26 / 94.43 37.73 / 94.19 35.03 / 90.32 34.22 / 91.77 38.03 / 93.21 36.84 / 94.55 38.98 / 94.11 37.33 / 95.28 zhs → en 39.59 / 97.18 36.13 / 95.04 38.26 / 97.28 38.12 / 97.38 34.82 / 97.52 32.44 / 95.14 30.84 / 95.57 35.40 / 96.59 33.74 / 96.79 36.60 / 97.21 34.94 / 97.38 en → zht 33.84 / 92.75 12.44 / 56.52 33.63 / 93.86 35.49 / 94.20 31.41 / 93.59 27.61 / 88.67 26.77 / 90.43 30.76 / 92.20 28.36 / 93.69 32.45 / 93.48 29.70 / 94.92 zht → en 37.95 / 96.15 32.27 / 92.00 37.02 / 96.88 37.13 / 97.20 33.52 / 96.86 30.77 / 93.61 29.61 / 94.05 33.21 / 95.81 31.98 / 95.87 34.77 / 96.42 33.43 / 96.71 en → da 55.61 / 97.25 50.04 / 95.66 53.61 / 98.02 54.62 / 98.18 51.23 / 97.82 49.11 / 95.10 47.78 / 95.63 52.57 / 97.14 50.55 / 97.60 54.70 / 97.81 51.72 / 98.31 da → en 58.03 / 97.39 53.07 / 95.43 54.84 / 97.40 55.88 / 97.52 53.48 / 97.23 53.32 / 95.92 52.19 / 96.14 55.01 / 97.23 53.94 / 97.43 55.72 / 97.45 54.44 / 97.49 en → de 49.24 / 98.81 46.62 / 97.50 49.54 / 98.55 51.94 / 98.67 47.64 / 98.12 43.10 / 96.45 40.74 / 96.93 47.32 / 98.28 45.88 / 98.45 48.91 / 98.50 46.04 / 98.83 de → en 52.41 / 97.24 49.80 / 96.06 50.31 / 97.16 51.05 / 97.31 49.52 / 97.12 47.82 / 95.87 46.73 / 95.99 49.91 / 96.84 48.51 / 97.05 51.28 / 97.20 49.43 / 97.33 en → el 42.95 / 93.61 38.73 / 91.03 42.78 / 95.01 43.62 / 94.86 41.10 / 94.26 35.76 / 87.18 35.47 / 88.87 40.61 / 93.29 39.40 / 94.34 42.02 / 94.31 41.05 / 95.52 el → en 45.33 / 94.85 43.70 / 93.79 44.00 / 95.12 43.51 / 95.45 41.68 / 95.03 38.94 / 91.92 37.33 / 92.47 42.19 / 94.64 40.84 / 94.97 43.52 / 95.33 42.16 / 95.46 en → tl 39.72 / 82.85 38.34 / 83.28 38.74 / 90.18 40.47 / 90.53 38.45 / 90.12 33.97 / 83.60 32.48 / 86.15 37.96 / 87.95 35.10 / 90.14 39.53 / 89.18 36.66 / 91.77 tl → en 58.10 / 94.74 54.58 / 92.86 56.49 / 95.01 56.90 / 95.08 52.61 / 95.00 49.06 / 91.45 47.64 / 91.94 53.41 / 94.21 51.93 / 94.57 55.41 / 94.92 53.80 / 95.10 en → fi 42.30 / 94.59 36.57 / 90.00 42.85 / 96.63 44.74 / 96.74 40.05 / 95.98 30.78 / 83.81 30.29 / 84.93 37.12 / 93.26 36.22 / 94.29 39.92 / 95.00 39.06 / 95.92 fi → en 44.26 / 95.41 41.22 / 92.87 42.56 / 95.72 43.23 / 96.07 41.05 / 95.78 37.82 / 92.00 37.09 / 92.64 41.12 / 95.09 40.01 / 95.13 42.35 / 95.80 40.91 / 95.80 en → fr 59.40 / 95.81 56.16 / 92.44 56.82 / 95.68 59.16 / 96.06 53.18 / 95.76 51.67 / 92.44 48.76 / 93.11 55.45 / 95.21 50.97 / 95.51 57.64 / 95.74 51.68 / 96.42 fr → en 54.39 / 96.33 51.54 / 95.11 51.46 / 96.39 52.26 / 96.77 49.82 / 96.45 48.37 / 95.58 47.24 / 95.80 50.40 / 96.60 48.67 / 96.58 51.60 / 96.71 50.23 / 96.90 en → he 51.65 / 93.63 46.82 / 90.79 48.01 / 94.89 51.50 / 95.15 45.42 / 94.02 37.22 / 83.68 35.53 / 84.96 45.45 / 92.33 41.66 / 93.60 48.52 / 94.52 43.61 / 95.36 he → en 53.64 / 95.84 49.02 / 92.85 52.37 / 96.63 52.61 / 96.92 49.84 / 96.10 45.33 / 91.03 43.95 / 91.47 49.92 / 95.20 48.54 / 95.18 51.57 / 96.09 49.95 / 96.11 en → hi 39.01 / 90.42 40.55 / 81.38 39.00 / 89.26 38.21 / 90.73 33.19 / 89.55 32.05 / 78.81 30.18 / 82.16 36.49 / 86.40 33.81 / 89.10 37.96 / 88.94 33.40 / 91.60 hi → en 51.39 / 94.34 47.26 / 93.16 49.87 / 95.01 49.21 / 95.09 46.42 / 95.02 42.33 / 90.16 41.41 / 90.77 46.43 / 93.52 45.47 / 93.89 47.95 / 94.27 47.29 / 94.69 en → hr 44.17 / 96.19 38.90 / 92.74 43.59 / 97.34 44.21 / 97.25 40.14 / 96.18 34.62 / 91.24 34.77 / 92.26 40.65 / 95.33 38.77 / 96.16 42.21 / 96.40 39.96 / 97.24 hr → en 47.39 / 96.26 42.66 / 93.89 46.02 / 95.95 46.20 / 96.16 44.88 / 95.96 42.42 / 94.13 41.58 / 94.60 44.98 / 95.77 44.42 / 96.06 45.83 / 96.18 44.94 / 96.50 en → hu 41.67 / 95.23 38.13 / 92.50 40.62 / 96.65 42.08 / 96.90 37.78 / 94.89 32.10 / 87.67 30.60 / 89.44 37.69 / 94.32 35.44 / 95.48 39.34 / 96.12 36.76 / 96.71 hu → en 46.39 / 95.33 41.90 / 92.66 44.29 / 95.87 44.30 / 95.92 42.16 / 95.61 39.76 / 92.23 38.75 / 92.82 42.93 / 95.06 41.59 / 95.24 44.76 / 95.58 43.08 / 95.83 en → id 53.26 / 96.16 49.18 / 92.62 48.92 / 96.85 50.69 / 96.67 44.63 / 96.75 47.59 / 94.39 43.42 / 95.25 49.38 / 96.33 43.45 / 97.04 49.68 / 96.90 42.23 / 97.56 id → en 53.92 / 95.90 49.88 / 93.73 53.54 / 96.27 53.70 / 96.49 50.49 / 96.24 48.03 / 94.54 47.91 / 94.84 51.18 / 96.00 50.49 / 96.01 52.23 / 96.29 51.56 / 96.63 en → it 41.80 / 96.42 38.27 / 94.18 40.20 / 96.68 41.44 / 96.62 38.35 / 96.33 35.87 / 93.61 34.14 / 94.25 38.64 / 95.80 36.89 / 96.74 39.94 / 96.48 38.57 / 97.27 it → en 43.13 / 96.48 41.56 / 94.71 40.66 / 96.60 40.72 / 96.84 39.13 / 96.25 37.93 / 95.25 37.03 / 95.60 39.72 / 96.40 39.13 / 96.73 40.56 / 96.69 39.68 / 96.86 en → ja 36.98 / 94.07 20.07 / 78.09 34.57 / 94.82 35.86 / 94.99 31.70 / 93.82 27.47 / 87.60 27.55 / 90.17 32.62 / 93.03 30.89 / 94.15 34.90 / 94.25 32.52 / 95.02 ja → en 38.14 / 95.48 34.54 / 91.67 36.69 / 95.94 36.98 / 96.26 33.65 / 95.78 30.79 / 91.51 29.55 / 92.38 33.66 / 94.77 32.39 / 95.29 35.36 / 95.65 33.86 / 96.21 en → k 38.93 / 83.31 34.04 / 80.51 34.53 / 89.87 35.84 / 89.84 31.82 / 88.12 27.55 / 73.45 26.82 / 75.58 33.87 / 85.74 32.69 / 86.98 35.65 / 88.80 32.55 / 89.80 k → en 45.50 / 85.42 33.68 / 69.63 44.17 / 87.18 44.56 / 87.60 42.38 / 87.45 35.22 / 78.90 34.92 / 80.73 40.10 / 84.76 39.12 / 85.47 42.12 / 86.49 41.51 / 87.15 en → km 27.42 / 74.81 22.97 / 71.55 27.21 / 81.05 28.15 / 82.72 23.39 / 78.37 21.48 / 69.74 21.46 / 72.98 25.35 / 79.86 25.27 / 81.82 27.18 / 81.91 27.12 / 84.13 km → en 43.74 / 80.19 38.64 / 71.55 42.85 / 82.09 43.67 / 82.19 36.65 / 82.32 34.09 / 75.35 32.00 / 77.05 38.38 / 80.68 36.05 / 81.90 40.34 / 82.17 37.32 / 83.48 en → ko 30.71 / 94.74 26.72 / 87.22 30.49 / 95.27 31.52 / 95.47 28.13 / 94.65 24.56 / 88.02 24.07 / 90.02 28.29 / 93.39 27.61 / 94.80 30.09 / 94.56 28.26 / 96.14 ko → en 39.67 / 94.60 35.42 / 91.21 39.89 / 95.55 39.80 / 95.91 37.17 / 95.50 33.52 / 91.22 32.52 / 91.93 36.76 / 93.87 35.88 / 94.19 39.01 / 94.87 37.92 / 95.19 en → lo 29.62 / 73.83 29.59 / 77.00 31.75 / 85.15 33.99 / 85.89 28.69 / 81.10 24.95 / 72.54 25.63 / 75.50 32.51 / 83.86 31.05 / 85.77 33.92 / 85.60 31.26 / 87.82 lo → en 46.52 / 87.43 42.40 / 81.17 47.69 / 88.51 47.70 / 89.12 41.83 / 89.00 37.59 / 80.91 35.70 / 81.86 42.74 / 86.46 41.32 / 87.26 45.29 / 87.78 43.39 / 88.52 en → my 24.43 / 68.53 17.74 / 68.43 26.45 / 81.22 26.94 / 82.36 19.67 / 76.21 13.20 / 59.40 16.41 / 63.19 23.37 / 77.46 22.38 / 79.39 26.14 / 81.59 23.99 / 83.76 my → en 38.37 / 80.02 34.73 / 74.94 39.72 / 83.35 40.26 / 83.59 33.04 / 82.78 27.66 / 68.92 26.91 / 71.52 33.76 / 78.76 32.80 / 80.31 36.22 / 80.94 34.68 / 82.38 en → nl 36.95 / 97.44 35.61 / 95.19 37.16 / 97.58 37.69 / 97.60 36.06 / 97.40 33.68 / 94.49 32.83 / 95.13 36.16 / 96.79 34.83 / 97.32 37.68 / 97.49 35.61 / 97.98 nl → en 40.30 / 96.35 38.94 / 94.89 38.75 / 96.45 38.65 / 96.51 37.93 / 96.41 36.89 / 94.95 35.91 / 95.11 38.12 / 96.29 37.58 / 96.47 39.16 / 96.50 38.13 / 96.75 en → nb 65.84 / 97.77 65.93 / 94.10 57.48 / 97.98 58.88 / 98.09 53.72 / 97.72 57.61 / 95.58 54.18 / 96.11 60.48 / 97.62 54.69 / 97.92 60.57 / 97.98 53.44 / 98.21 nb → en 74.06 / 97.55 53.90 / 83.91 70.17 / 97.31 70.92 / 97.30 67.68 / 97.10 66.82 / 95.43 65.99 / 95.58 69.30 / 96.86 68.40 / 96.96 70.65 / 97.36 69.48 / 97.49 en → fa 39.73 / 93.68 36.13 / 88.66 35.27 / 94.20 36.86 / 94.47 31.54 / 93.52 29.57 / 83.00 28.13 / 85.85 34.06 / 91.71 31.24 / 93.17 36.08 / 93.57 31.65 / 94.66 fa → en 46.99 / 93.97 44.19 / 91.27 46.00 / 94.73 45.90 / 94.83 42.58 / 95.11 38.40 / 89.26 37.00 / 90.25 42.91 / 92.89 40.80 / 93.24 44.66 / 94.03 43.15 / 94.36 en → pl 36.10 / 96.35 32.54 / 91.17 36.63 / 96.65 37.39 / 96.71 35.73 / 96.54 30.08 / 88.55 29.34 / 90.07 33.39 / 94.47 32.13 / 95.45 35.58 / 95.91 33.77 / 96.70 pl → en 39.14 / 95.58 36.68 / 93.35 37.81 / 95.83 37.92 / 96.10 36.31 / 95.53 34.03 / 93.76 32.74 / 94.11 36.60 / 95.52 35.13 / 95.79 37.72 / 95.97 35.98 / 96.33 en → pt 58.08 / 96.96 52.87 / 95.08 53.47 / 97.01 57.15 / 97.09 52.59 / 97.02 51.98 / 95.44 49.86 / 96.04 54.31 / 96.82 50.14 / 97.41 55.03 / 97.44 50.04 / 97.87 pt → en 57.81 / 96.71 55.19 / 95.31 56.25 / 96.72 56.85 / 96.93 54.93 / 96.48 53.01 / 95.46 52.47 / 95.87 55.68 / 96.56 54.37 / 96.58 56.65 / 96.67 55.48 / 96.86 en → ro 51.54 / 95.10 44.66 / 91.48 51.00 / 96.32 52.24 / 96.44 49.34 / 95.61 45.43 / 90.87 44.43 / 92.10 49.18 / 94.67 47.72 / 95.84 50.74 / 95.78 48.69 / 96.73 ro → en 52.64 / 95.89 50.80 / 94.29 49.77 / 96.12 51.03 / 96.48 48.99 / 96.14 47.25 / 94.27 45.74 / 94.87 49.42 / 96.00 47.39 / 96.06 50.99 / 96.41 48.74 / 96.61 en → ru 43.93 / 95.90 41.03 / 92.59 43.22 / 96.24 44.33 / 96.34 41.49 / 95.49 36.55 / 89.66 35.70 / 90.83 40.74 / 94.24 39.07 / 94.95 42.60 / 95.41 39.70 / 96.40 ru → en 45.46 / 95.87 42.21 / 94.18 43.32 / 96.30 43.31 / 96.33 41.32 / 96.31 38.46 / 94.15 37.54 / 94.35 41.92 / 95.92 40.72 / 96.20 42.66 / 96.26 42.18 / 96.59 en → sk 48.02 / 94.98 42.95 / 91.52 45.41 / 95.95 47.22 / 96.01 43.34 / 95.66 38.50 / 88.73 37.35 / 90.43 42.16 / 94.04 41.59 / 95.07 45.21 / 95.37 42.89 / 96.29 sk → en 50.67 / 96.02 45.21 / 93.06 47.68 / 96.07 48.45 / 96.42 46.03 / 95.80 43.60 / 93.58 42.02 / 94.19 46.57 / 95.85 44.61 / 96.01 47.98 / 96.24 46.06 / 96.46 en → sl 43.79 / 95.57 38.15 / 91.54 42.09 / 96.96 44.41 / 97.16 41.23 / 95.89 34.28 / 89.03 33.33 / 90.47 39.45 / 94.48 38.50 / 95.17 41.35 / 96.02 39.67 / 96.83 sl → en 45.04 / 95.61 40.93 / 93.06 43.27 / 96.01 44.22 / 96.23 42.22 / 96.01 39.16 / 93.46 38.42 / 93.95 42.30 / 95.52 41.21 / 95.72 43.62 / 96.03 42.49 / 96.13 en → es 35.23 / 96.69 33.12 / 94.94 34.71 / 96.97 35.23 / 96.71 34.81 / 96.71 32.33 / 95.09 32.19 / 95.94 34.29 / 96.43 34.27 / 97.06 34.75 / 96.98 34.62 / 97.50 es → en 39.54 / 95.79 35.43 / 92.64 37.61 / 96.21 36.97 / 96.30 37.06 / 96.15 34.95 / 95.03 34.15 / 95.43 36.74 / 96.00 35.37 / 96.07 37.45 / 96.26 36.13 / 96.38 en → sv 55.56 / 97.32 50.11 / 95.51 53.32 / 98.18 54.61 / 98.41 50.35 / 97.74 48.31 / 94.81 46.89 / 95.56 51.83 / 97.16 49.68 / 97.59 53.51 / 97.98 50.46 / 98.43 sv → en 56.87 / 96.69 52.41 / 95.20 55.39 / 96.77 56.02 / 96.90 53.80 / 96.79 52.91 / 95.27 51.61 / 95.36 54.60 / 96.55 53.57 / 96.64 55.53 / 96.81 53.85 / 96.94 en → ta 39.75 / 80.24 36.56 / 76.17 34.03 / 84.98 32.47 / 86.88 26.73 / 82.90 24.68 / 71.62 23.55 / 75.46 31.75 / 82.24 28.61 / 85.05 33.51 / 83.99 28.93 / 87.38 ta → en 43.72 / 90.26 40.64 / 87.06 43.37 / 91.45 42.87 / 91.70 40.17 / 91.78 33.17 / 81.57 32.61 / 82.49 38.22 / 88.21 37.56 / 89.02 40.87 / 90.41 39.92 / 91.22 en → th 45.55 / 93.10 35.08 / 83.17 45.41 / 95.08 46.53 / 95.92 44.36 / 94.82 37.69 / 88.13 37.27 / 90.16 42.50 / 93.62 40.01 / 94.79 44.95 / 94.72 41.47 / 96.12 th → en 40.42 / 94.67 36.87 / 90.61 42.37 / 95.85 43.05 / 96.10 38.42 / 95.64 34.55 / 91.78 32.87 / 92.54 38.73 / 94.69 36.94 / 95.24 40.37 / 95.79 38.73 / 95.93 en → tr 41.72 / 93.54 41.52 / 89.15 43.49 / 94.17 45.49 / 94.55 39.63 / 93.87 35.98 / 86.34 34.78 / 88.41 40.55 / 92.03 37.96 / 93.60 42.55 / 94.19 38.15 / 95.55 tr → en 49.43 / 94.41 45.82 / 92.28 48.16 / 95.39 47.91 / 95.39 45.02 / 95.16 42.25 / 91.77 41.38 / 92.16 46.06 / 94.39 45.28 / 94.71 47.58 / 95.23 46.16 / 95.55 en → ur 32.60 / 79.75 30.52 / 78.50 30.90 / 90.56 31.29 / 91.34 28.30 / 88.86 23.39 / 76.97 24.50 / 81.20 30.08 / 86.66 27.74 / 89.32 31.37 / 89.26 28.08 / 91.68 ur → en 45.50 / 91.56 43.08 / 88.91 44.86 / 93.43 44.41 / 93.64 41.28 / 93.53 35.95 / 84.67 34.75 / 85.79 41.20 / 91.08 39.95 / 91.37 43.45 / 92.45 41.45 / 92.95 en → uz 37.86 / 81.27 29.97 / 81.96 32.01 / 88.85 34.40 / 89.14 29.82 / 86.79 24.86 / 72.20 23.02 / 75.36 31.26 / 84.02 27.27 / 85.97 33.64 / 86.89 28.23 / 88.60 uz → en 46.29 / 85.29 39.47 / 78.22 44.68 / 87.07 45.02 / 87.70 41.78 / 87.62 36.21 / 78.28 35.22 / 80.17 40.79 / 84.67 39.71 / 85.45 43.69 / 86.52 42.18 / 87.10 en → vi 46.35 / 94.54 43.30 / 88.37 45.62 / 94.96 46.16 / 95.34 42.39 / 94.93 41.77 / 91.32 40.05 / 92.59 44.46 / 93.69 41.20 / 95.13 45.33 / 94.68 41.13 / 96.16 vi → en 46.81 / 95.45 43.75 / 93.72 45.53 / 96.00 46.07 / 96.29 42.75 / 96.06 40.81 / 93.57 39.53 / 93.84 43.83 / 95.32 42.44 / 95.56 44.91 / 95.67 43.60 / 96.11 en → yue 36.05 / 83.12 23.24 / 78.83 35.19 / 90.20 35.91 / 89.75 32.93 / 90.15 30.56 / 85.35 30.43 / 86.58 34.70 / 89.50 33.27 / 90.92 36.73 / 90.85 34.13 / 91.99 yue → en 35.41 / 92.51 27.43 / 84.53 38.98 / 96.45 39.23 / 96.62 35.71 / 96.22 32.62 / 92.40 31.42 / 92.77 35.85 / 95.05 34.42 / 95.24 37.58 / 95.79 35.93 / 95.95 en → ms 47.92 / 91.69 45.54 / 90.47 43.06 / 95.41 44.56 / 95.27 39.99 / 95.65 41.11 / 93.39 38.73 / 94.54 44.07 / 95.08 38.91 / 96.05 45.24 / 95.67 37.83 / 96.94 ms → en 54.47 / 95.89 51.20 / 93.58 53.47 / 96.29 53.69 / 96.42 51.14 / 96.04 48.07 / 94.10 47.62 / 94.36 51.32 / 95.55 50.82 / 95.89 53.02 / 96.06 52.31 / 96.17 Table 13: Reference-based English-centric evaluation results (spBLEU / XCOMET) of baseline models (Part I) and MiLMMT models on the FLORES+ benchmark. Direction TranslateGemma-4B TranslateGemma-12B TranslateGemma-27B GemmaX2-28-2B GemmaX2-28-9B Hunyuan-MT-7B HY-MT1.5-1.8B HY-MT1.5-7B HY-MT2-1.8B HY-MT2-7B HY-MT2-30B-A3B Seed-X-Instruct-7B Seed-X-PPO-7B Tower-Plus-2B Tower-Plus-9B Tower-Plus-72B zhs → ar 20.57 / 85.48 22.40 / 90.53 23.94 / 92.21 23.78 / 85.41 28.27 / 90.53 23.10 / 91.34 20.05 / 86.54 22.37 / 91.54 22.22 / 87.65 24.07 / 92.15 26.50 / 92.09 26.36 / 88.77 26.77 / 90.74 - / - - / - - / - ar → zhs 23.42 / 81.50 26.12 / 87.23 28.87 / 88.59 29.88 / 83.94 32.97 / 87.55 22.93 / 88.19 22.40 / 84.78 24.11 / 88.83 23.76 / 84.67 24.10 / 89.41 30.56 / 90.10 31.85 / 86.49 27.59 / 87.27 - / - - / - - / - zhs → az 4.75 / 34.69 15.04 / 71.10 17.28 / 79.83 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - az → zhs 18.51 / 68.43 21.47 / 78.90 23.13 / 81.09 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → bn 16.84 / 83.05 18.08 / 88.10 20.66 / 89.69 20.33 / 84.02 23.41 / 88.28 18.16 / 87.60 16.49 / 83.25 17.86 / 88.29 17.70 / 84.48 18.86 / 90.62 20.53 / 91.40 - / - - / - - / - - / - - / - bn → zhs 21.35 / 74.92 24.93 / 81.79 26.90 / 83.97 26.94 / 78.04 30.01 / 82.77 18.29 / 77.93 18.56 / 76.90 20.32 / 80.24 19.01 / 76.65 20.88 / 83.59 27.33 / 85.93 - / - - / - - / - - / - - / - zhs → bg 23.72 / 87.94 25.90 / 93.10 28.28 / 94.30 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - bg → zhs 26.16 / 84.38 28.24 / 87.89 30.11 / 88.89 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → ca 21.71 / 84.58 25.94 / 90.70 24.19 / 92.08 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - ca → zhs 26.10 / 81.89 28.91 / 87.52 30.81 / 88.52 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → cs 19.44 / 79.11 22.18 / 87.18 23.52 / 88.99 25.40 / 84.69 28.55 / 89.72 23.53 / 88.14 20.62 / 80.47 22.57 / 88.11 24.81 / 83.93 28.24 / 89.99 28.96 / 91.21 26.13 / 85.35 28.68 / 89.67 23.50 / 82.00 27.47 / 88.88 28.39 / 89.56 cs → zhs 25.91 / 80.51 27.66 / 84.90 29.77 / 86.22 31.14 / 82.51 34.05 / 85.17 24.99 / 87.15 24.04 / 84.15 25.20 / 87.15 26.91 / 83.76 26.93 / 88.42 33.24 / 87.18 32.84 / 82.62 30.17 / 84.50 29.81 / 78.89 33.81 / 83.99 36.21 / 84.79 zhs → zht 11.59 / 94.06 19.39 / 95.16 29.06 / 96.05 - / - - / - 25.22 / 95.35 23.86 / 94.62 25.86 / 95.60 28.17 / 95.29 26.96 / 96.15 29.47 / 94.65 - / - - / - 30.81 / 94.26 26.00 / 94.27 29.78 / 94.57 zht → zhs 10.84 / 94.19 31.17 / 95.18 32.75 / 95.39 - / - - / - 28.87 / 95.83 28.03 / 94.65 28.33 / 95.71 30.76 / 93.62 31.03 / 94.45 29.55 / 92.43 - / - - / - 33.23 / 93.65 34.43 / 94.28 34.35 / 94.01 zhs → da 22.74 / 91.40 25.52 / 94.77 26.23 / 95.83 - / - - / - - / - - / - - / - - / - - / - - / - 29.00 / 93.91 30.91 / 95.58 25.54 / 91.99 28.98 / 94.68 29.85 / 95.07 da → zhs 26.93 / 87.00 29.11 / 90.98 30.93 / 91.69 - / - - / - - / - - / - - / - - / - - / - - / - 33.99 / 89.84 31.16 / 91.23 30.70 / 86.48 34.94 / 89.98 37.03 / 90.85 zhs → de 22.32 / 95.26 24.66 / 97.34 26.45 / 97.78 28.04 / 96.11 30.67 / 97.39 25.36 / 97.37 20.03 / 94.66 24.76 / 97.36 23.12 / 95.98 25.67 / 97.48 28.37 / 98.05 29.03 / 97.08 30.82 / 97.61 24.82 / 95.31 29.04 / 97.23 30.08 / 97.46 de → zhs 26.79 / 87.69 28.18 / 91.89 29.80 / 92.57 32.39 / 89.90 34.64 / 91.81 26.32 / 93.15 25.73 / 91.29 25.85 / 93.26 28.85 / 90.84 28.55 / 93.43 31.23 / 93.94 33.54 / 91.04 29.53 / 92.03 30.39 / 87.43 34.49 / 91.10 36.15 / 92.05 zhs → el 20.10 / 81.89 24.24 / 89.95 26.17 / 91.70 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - el → zhs 22.79 / 79.35 25.82 / 84.52 27.94 / 85.51 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → en 27.24 / 95.63 28.90 / 96.71 30.94 / 97.27 34.68 / 96.48 36.57 / 96.96 29.57 / 96.86 23.89 / 96.00 27.48 / 96.82 26.95 / 96.17 28.98 / 97.04 29.82 / 97.68 36.47 / 96.46 33.19 / 96.86 32.02 / 95.53 34.78 / 96.77 36.72 / 97.26 en → zhs 31.02 / 91.99 31.96 / 94.45 33.60 / 95.15 37.94 / 92.88 39.85 / 93.72 31.03 / 96.04 30.57 / 93.81 29.94 / 95.58 35.79 / 93.47 36.01 / 95.64 37.68 / 95.71 37.60 / 94.01 34.09 / 95.28 36.39 / 90.84 39.92 / 93.32 42.21 / 94.51 zhs → tl 18.44 / 82.16 20.95 / 86.47 23.21 / 87.89 23.44 / 83.91 25.74 / 86.88 17.81 / 79.00 16.72 / 80.67 17.21 / 79.27 20.90 / 83.77 24.23 / 88.14 25.07 / 88.47 - / - - / - - / - - / - - / - tl → zhs 25.73 / 74.28 27.41 / 80.25 29.86 / 81.63 31.42 / 77.91 34.20 / 80.73 23.22 / 80.73 23.17 / 77.04 23.82 / 80.99 26.39 / 77.64 25.74 / 83.34 32.49 / 83.16 - / - - / - - / - - / - - / - zhs → fi 17.40 / 77.32 21.62 / 87.49 23.32 / 90.29 - / - - / - - / - - / - - / - - / - - / - - / - 27.87 / 85.96 29.19 / 90.04 20.57 / 81.75 25.33 / 89.53 24.80 / 88.02 fi → zhs 24.45 / 80.16 26.60 / 86.33 28.57 / 87.24 - / - - / - - / - - / - - / - - / - - / - - / - 31.45 / 82.90 28.18 / 84.84 27.73 / 79.77 32.30 / 80.85 33.81 / 85.54 zhs → fr 27.86 / 84.17 30.30 / 88.21 29.75 / 89.78 33.99 / 87.57 36.76 / 90.82 29.40 / 89.32 24.35 / 85.62 28.54 / 89.34 29.19 / 86.69 32.01 / 90.36 33.69 / 91.65 36.43 / 89.25 35.23 / 91.02 30.17 / 84.99 33.70 / 89.90 35.61 / 91.42 fr → zhs 26.80 / 85.25 28.26 / 88.69 29.81 / 89.42 32.53 / 86.39 34.74 / 87.88 26.36 / 90.78 25.47 / 87.64 25.98 / 90.87 29.10 / 87.28 28.36 / 91.08 32.62 / 90.50 33.71 / 86.59 30.29 / 88.72 30.27 / 83.46 34.26 / 87.35 36.22 / 88.39 zhs → he 17.50 / 78.51 22.43 / 89.71 25.07 / 91.72 25.00 / 84.69 28.36 / 90.70 18.51 / 81.85 8.12 / 74.47 19.08 / 83.97 19.57 / 82.23 22.87 / 91.40 25.92 / 92.13 - / - - / - - / - - / - - / - he → zhs 24.27 / 82.54 27.10 / 88.67 30.14 / 90.15 30.68 / 86.11 33.60 / 89.02 22.40 / 87.55 21.81 / 83.82 23.40 / 88.08 23.75 / 83.98 24.23 / 90.43 29.09 / 91.74 - / - - / - - / - - / - - / - zhs → hi 18.60 / 83.28 20.12 / 88.39 22.34 / 89.60 21.78 / 81.79 24.86 / 86.97 18.23 / 88.38 15.90 / 83.52 17.19 / 88.22 17.56 / 84.21 19.40 / 90.41 20.81 / 91.03 - / - - / - 18.77 / 76.83 23.07 / 85.13 23.46 / 83.67 hi → zhs 24.06 / 80.91 26.51 / 86.32 28.71 / 88.22 29.08 / 82.58 31.76 / 85.68 20.61 / 85.69 21.39 / 82.43 22.53 / 87.14 23.17 / 81.89 23.42 / 88.77 29.05 / 89.93 - / - - / - 26.37 / 79.27 31.76 / 85.20 33.37 / 85.57 zhs → hr 16.42 / 82.55 19.95 / 91.60 22.86 / 93.62 - / - - / - - / - - / - - / - - / - - / - - / - 24.18 / 86.95 26.80 / 91.49 - / - - / - - / - hr → zhs 25.26 / 83.55 27.67 / 88.47 29.49 / 89.76 - / - - / - - / - - / - - / - - / - - / - - / - 31.68 / 86.07 29.49 / 87.44 - / - - / - - / - zhs → hu 16.75 / 77.55 20.67 / 90.08 22.70 / 92.83 - / - - / - - / - - / - - / - - / - - / - - / - 24.77 / 90.34 27.12 / 93.02 21.90 / 87.07 26.31 / 92.80 25.35 / 92.19 hu → zhs 24.42 / 82.70 26.40 / 88.36 28.69 / 89.37 - / - - / - - / - - / - - / - - / - - / - - / - 32.13 / 86.64 29.45 / 88.26 28.53 / 83.60 32.76 / 87.46 34.26 / 88.13 zhs → id 22.91 / 92.62 24.63 / 94.89 25.67 / 95.29 29.24 / 93.50 30.68 / 94.85 22.67 / 94.82 19.48 / 92.54 21.57 / 94.88 21.60 / 93.08 22.75 / 95.60 26.21 / 95.54 28.87 / 93.17 30.21 / 94.75 - / - - / - - / - id → zhs 26.84 / 87.08 28.99 / 91.04 30.71 / 91.63 32.77 / 89.39 34.59 / 90.99 25.92 / 91.99 24.95 / 89.58 25.79 / 91.90 28.10 / 90.00 28.06 / 92.87 32.80 / 92.58 34.33 / 90.49 30.93 / 91.51 - / - - / - - / - zhs → it 22.06 / 89.17 24.76 / 92.92 23.76 / 93.47 25.68 / 90.65 28.65 / 93.26 22.01 / 92.81 18.96 / 89.99 21.59 / 92.86 20.09 / 90.47 22.38 / 93.73 24.37 / 94.48 28.01 / 92.50 28.37 / 93.84 22.43 / 88.32 27.43 / 92.72 28.57 / 93.90 it → zhs 23.09 / 86.41 25.85 / 90.57 27.41 / 91.46 30.05 / 88.74 32.28 / 90.40 24.26 / 92.14 23.51 / 89.79 24.17 / 92.31 26.19 / 89.03 25.75 / 92.71 29.16 / 92.57 31.92 / 88.48 29.04 / 89.78 28.58 / 86.01 32.09 / 89.34 34.25 / 90.53 zhs → ja 20.17 / 83.63 21.75 / 88.59 22.67 / 88.46 23.27 / 87.63 25.33 / 89.46 22.89 / 90.95 20.22 / 88.05 22.34 / 90.53 22.54 / 88.37 25.25 / 90.97 25.90 / 91.77 28.13 / 86.98 29.39 / 90.28 23.39 / 86.50 26.39 / 89.65 27.49 / 90.40 ja → zhs 23.67 / 81.76 24.77 / 85.68 26.37 / 87.24 27.67 / 83.26 29.68 / 85.62 22.72 / 88.04 21.59 / 85.01 23.10 / 88.26 24.43 / 84.71 25.40 / 87.86 28.30 / 87.85 29.62 / 83.14 27.26 / 84.61 26.80 / 81.01 29.53 / 84.51 30.30 / 85.02 zhs → k 6.77 / 36.44 16.04 / 64.31 20.27 / 74.62 - / - - / - 0.19 / 26.58 0.23 / 23.79 0.11 / 26.25 0.22 / 23.98 0.15 / 24.41 0.35 / 28.54 - / - - / - - / - - / - - / - k → zhs 21.24 / 66.30 25.30 / 77.82 27.74 / 79.50 - / - - / - 13.56 / 49.20 11.28 / 31.27 15.81 / 52.16 11.35 / 32.06 18.98 / 64.59 28.22 / 79.06 - / - - / - - / - - / - - / - zhs → km 3.35 / 30.53 12.04 / 56.95 16.57 / 68.55 19.66 / 70.30 22.48 / 76.83 14.82 / 63.19 0.15 / 56.67 14.97 / 65.09 19.70 / 72.98 20.19 / 78.67 22.18 / 79.77 - / - - / - - / - - / - - / - km → zhs 17.30 / 60.44 22.14 / 72.05 24.27 / 74.20 25.75 / 70.87 29.68 / 75.63 16.59 / 67.78 19.29 / 71.51 18.70 / 71.03 19.81 / 72.76 20.22 / 77.41 24.88 / 78.79 - / - - / - - / - - / - - / - zhs → ko 17.78 / 88.27 19.47 / 92.56 20.72 / 93.84 19.91 / 89.81 22.76 / 92.86 20.11 / 93.47 16.93 / 90.04 19.40 / 93.27 19.84 / 91.15 20.38 / 93.78 21.80 / 94.23 18.02 / 85.05 20.68 / 88.94 18.98 / 88.37 22.53 / 92.84 23.22 / 92.94 ko → zhs 23.85 / 81.21 26.09 / 85.69 27.56 / 86.83 28.73 / 82.93 31.48 / 85.45 24.03 / 87.28 23.21 / 84.57 23.88 / 87.57 25.36 / 84.37 24.76 / 88.36 29.08 / 88.08 30.12 / 82.84 27.27 / 84.26 27.28 / 80.07 30.77 / 84.58 32.05 / 85.64 zhs → lo 2.12 / 35.46 13.90 / 67.00 19.65 / 76.70 21.09 / 74.01 25.18 / 81.44 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - lo → zhs 18.14 / 63.25 23.69 / 75.15 26.37 / 78.76 26.46 / 73.88 30.79 / 79.42 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → my 3.39 / 31.67 9.36 / 50.20 13.55 / 64.05 16.36 / 64.50 20.59 / 75.29 14.51 / 59.56 13.53 / 59.83 14.60 / 62.62 19.56 / 71.15 21.49 / 77.50 21.32 / 79.05 - / - - / - - / - - / - - / - my → zhs 13.91 / 46.51 19.24 / 58.03 21.70 / 62.36 21.79 / 58.20 26.20 / 65.69 13.01 / 52.33 16.91 / 57.39 15.88 / 57.54 17.82 / 60.76 18.11 / 66.87 24.67 / 67.64 - / - - / - - / - - / - - / - zhs → nl 19.52 / 90.71 21.72 / 94.15 22.63 / 95.18 23.87 / 92.74 25.78 / 95.06 19.82 / 92.82 17.14 / 89.39 20.02 / 92.97 19.52 / 91.13 21.55 / 94.73 23.40 / 96.03 28.83 / 94.09 30.01 / 95.49 22.06 / 91.08 25.19 / 94.62 26.38 / 95.07 nl → zhs 24.09 / 85.87 25.57 / 89.68 26.80 / 90.88 28.77 / 88.26 29.93 / 89.93 22.78 / 91.12 22.18 / 88.68 22.74 / 91.13 25.18 / 88.46 24.93 / 92.30 27.79 / 92.67 30.27 / 88.33 28.47 / 90.10 26.81 / 85.36 30.19 / 88.91 32.50 / 89.84 zhs → nb 21.19 / 91.91 24.05 / 94.96 24.88 / 95.89 - / - - / - - / - - / - - / - - / - - / - - / - 22.43 / 90.31 24.69 / 93.31 25.46 / 91.62 29.47 / 93.96 29.31 / 94.49 nb → zhs 28.50 / 86.21 30.25 / 90.25 32.29 / 91.26 - / - - / - - / - - / - - / - - / - - / - - / - 34.73 / 88.58 31.57 / 89.35 31.94 / 84.97 36.28 / 89.04 38.63 / 89.62 zhs → fa 17.23 / 84.34 19.16 / 90.73 20.58 / 92.14 22.96 / 87.19 24.90 / 91.09 19.21 / 87.86 17.00 / 83.10 18.82 / 89.37 18.39 / 85.82 19.63 / 91.51 22.10 / 92.35 - / - - / - - / - - / - - / - fa → zhs 23.20 / 82.82 25.69 / 87.91 28.13 / 89.58 29.38 / 85.42 32.25 / 88.47 21.42 / 87.43 20.68 / 83.27 22.54 / 88.47 21.74 / 83.58 22.69 / 89.73 29.41 / 90.48 - / - - / - - / - - / - - / - zhs → pl 18.16 / 84.27 21.18 / 90.07 22.61 / 91.98 23.96 / 88.58 26.08 / 92.12 18.69 / 85.97 18.09 / 82.45 18.67 / 87.79 22.46 / 85.89 25.15 / 91.89 26.48 / 93.32 27.27 / 88.96 29.11 / 92.17 21.37 / 86.27 25.78 / 92.11 25.42 / 92.35 pl → zhs 24.01 / 82.84 25.29 / 86.75 27.20 / 87.80 28.20 / 83.54 30.47 / 86.01 22.87 / 87.81 21.60 / 84.14 23.23 / 88.15 24.23 / 84.55 24.95 / 89.58 27.97 / 89.78 30.08 / 83.11 28.83 / 84.82 27.03 / 81.52 30.57 / 85.15 32.25 / 86.17 zhs → pt 24.94 / 92.31 26.88 / 94.40 25.45 / 94.86 30.48 / 93.25 32.52 / 94.96 26.28 / 94.47 22.73 / 92.99 25.95 / 94.84 25.68 / 93.10 26.89 / 95.08 29.61 / 95.75 32.86 / 93.86 28.30 / 95.13 25.09 / 90.94 27.96 / 94.01 29.57 / 95.03 pt → zhs 27.28 / 88.20 28.62 / 91.70 30.28 / 92.52 33.26 / 89.71 35.36 / 91.09 26.49 / 93.17 25.35 / 90.70 26.45 / 93.13 28.99 / 90.64 28.60 / 93.44 31.03 / 94.26 34.07 / 91.25 29.57 / 91.99 30.75 / 87.55 34.43 / 90.63 36.69 / 91.92 zhs → ro 23.83 / 83.55 26.01 / 88.59 27.45 / 90.98 - / - - / - - / - - / - - / - - / - - / - - / - 33.55 / 86.99 35.64 / 90.91 26.79 / 84.96 30.38 / 90.68 30.88 / 90.51 ro → zhs 26.24 / 83.15 28.28 / 87.22 30.66 / 88.21 - / - - / - - / - - / - - / - - / - - / - - / - 33.66 / 85.04 30.60 / 87.09 31.06 / 81.83 34.70 / 84.95 36.53 / 86.23 zhs → ru 21.68 / 90.79 24.27 / 93.95 24.30 / 94.62 26.24 / 90.35 28.54 / 93.46 23.09 / 93.52 19.98 / 89.63 22.65 / 93.45 22.19 / 90.45 24.53 / 94.02 27.83 / 94.83 28.83 / 93.54 29.17 / 94.43 23.76 / 90.00 27.89 / 93.44 29.34 / 94.11 ru → zhs 24.81 / 87.96 26.99 / 91.60 28.12 / 92.40 30.37 / 88.90 33.04 / 91.32 24.67 / 92.81 23.31 / 90.48 24.54 / 92.94 25.87 / 90.12 25.45 / 93.57 29.07 / 93.52 31.81 / 91.11 28.01 / 92.05 28.73 / 87.62 32.80 / 90.44 34.75 / 91.45 zhs → sk 16.35 / 76.64 20.16 / 87.29 21.89 / 90.02 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - sk → zhs 25.15 / 80.91 27.44 / 86.37 29.25 / 87.42 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → sl 14.09 / 70.55 19.07 / 86.90 21.47 / 90.16 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - sl → zhs 24.00 / 79.82 26.64 / 85.96 28.65 / 87.41 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → es 21.43 / 91.70 22.94 / 93.55 24.30 / 94.76 23.87 / 93.25 25.34 / 94.66 22.57 / 94.14 19.29 / 92.55 21.76 / 94.27 21.50 / 92.79 22.46 / 95.12 25.04 / 95.41 25.84 / 94.02 26.79 / 95.01 22.17 / 91.40 24.85 / 94.28 25.57 / 95.08 es → zhs 23.67 / 87.66 25.43 / 91.09 26.88 / 92.30 28.91 / 89.30 31.19 / 91.25 23.63 / 92.99 22.97 / 90.76 23.13 / 92.92 25.80 / 90.11 25.08 / 93.37 28.53 / 93.26 31.33 / 90.55 26.89 / 91.60 27.67 / 86.95 30.79 / 90.18 32.86 / 91.91 zhs → sv 22.89 / 91.31 24.81 / 95.07 26.10 / 96.29 - / - - / - - / - - / - - / - - / - - / - - / - 31.31 / 94.30 33.47 / 95.83 25.76 / 92.10 29.25 / 95.42 30.19 / 95.28 sv → zhs 26.62 / 87.63 28.64 / 91.25 30.31 / 92.09 - / - - / - - / - - / - - / - - / - - / - - / - 34.80 / 90.72 31.29 / 91.41 30.44 / 86.67 34.24 / 90.55 36.69 / 91.20 zhs → ta 19.39 / 77.02 19.95 / 83.10 21.41 / 84.30 - / - - / - 15.50 / 72.07 16.31 / 72.32 15.69 / 74.05 17.99 / 76.75 19.26 / 85.06 21.03 / 85.14 - / - - / - - / - - / - - / - ta → zhs 20.13 / 68.75 23.28 / 77.27 25.64 / 79.39 - / - - / - 16.08 / 71.78 17.05 / 69.82 18.37 / 75.23 16.49 / 69.87 18.61 / 79.87 26.16 / 80.43 - / - - / - - / - - / - - / - zhs → th 28.80 / 84.82 29.82 / 90.93 32.51 / 92.39 33.53 / 87.59 36.59 / 91.41 29.70 / 91.17 24.92 / 84.63 29.17 / 91.02 30.42 / 88.18 30.74 / 93.05 33.53 / 93.28 33.67 / 87.19 35.99 / 89.95 - / - - / - - / - th → zhs 22.97 / 83.58 25.24 / 88.30 26.75 / 89.73 28.96 / 86.31 31.82 / 89.22 21.54 / 88.80 20.90 / 85.84 22.06 / 88.91 22.58 / 85.83 22.43 / 90.53 29.07 / 90.83 30.23 / 86.44 26.42 / 87.33 - / - - / - - / - zhs → tr 18.36 / 80.69 21.07 / 87.47 22.94 / 89.86 24.32 / 84.72 27.04 / 88.76 21.62 / 89.26 18.18 / 84.99 20.49 / 88.92 21.00 / 85.12 23.66 / 89.28 23.99 / 90.96 27.50 / 84.15 28.59 / 87.93 - / - - / - - / - tr → zhs 24.53 / 80.73 26.68 / 86.57 29.42 / 87.79 30.37 / 83.90 32.85 / 86.63 24.23 / 87.43 23.76 / 84.91 24.61 / 87.74 25.81 / 84.77 26.49 / 89.17 30.95 / 89.13 31.74 / 84.72 26.69 / 85.94 - / - - / - - / - zhs → ur 9.10 / 67.15 15.13 / 83.76 17.76 / 87.14 18.34 / 79.92 21.47 / 86.57 13.73 / 79.50 12.74 / 75.83 14.33 / 81.25 15.71 / 80.42 17.42 / 88.92 19.17 / 89.18 - / - - / - - / - - / - - / - ur → zhs 20.46 / 70.90 24.43 / 80.44 26.72 / 82.74 26.58 / 76.19 30.48 / 81.23 17.90 / 77.40 17.66 / 71.49 20.60 / 79.76 18.23 / 71.70 21.35 / 83.16 27.38 / 85.05 - / - - / - - / - - / - - / - zhs → uz 0.13 / 19.92 15.06 / 65.85 15.84 / 74.53 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - uz → zhs 20.69 / 59.47 25.01 / 71.59 27.81 / 74.68 - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - - / - zhs → vi 25.66 / 90.15 27.03 / 93.29 28.54 / 94.25 31.93 / 91.97 33.15 / 93.25 26.15 / 93.96 23.25 / 91.39 25.02 / 93.71 26.18 / 92.14 25.88 / 94.42 28.73 / 94.74 32.55 / 90.50 34.40 / 92.92 - / - - / - - / - vi → zhs 24.52 / 86.98 26.91 / 91.13 29.43 / 92.05 30.44 / 89.26 32.72 / 90.90 24.82 / 91.99 23.77 / 90.48 24.41 / 92.19 25.74 / 90.23 25.40 / 92.91 30.79 / 92.74 32.46 / 90.07 29.16 / 91.18 - / - - / - - / - zhs → yue 13.78 / 94.07 21.49 / 95.42 23.72 / 95.69 - / - - / - 20.70 / 92.15 23.39 / 93.03 28.54 / 92.22 23.95 / 89.48 35.20 / 92.28 15.41 / 90.03 - / - - / - - / - - / - - / - yue → zhs 12.44 / 93.58 35.65 / 94.92 36.62 / 95.21 - / - - / - 32.81 / 95.77 33.40 / 94.17 32.72 / 95.94 36.11 / 93.20 36.40 / 93.86 36.52 / 93.75 - / - - / - - / - - / - - / - zhs → ms 17.68 / 87.43 20.28 / 92.82 22.91 / 93.68 25.56 / 92.39 26.77 / 93.76 21.65 / 93.09 15.58 / 90.79 19.70 / 93.65 20.45 / 92.21 20.91 / 94.81 23.66 / 94.69 27.14 / 91.00 27.77 / 93.03 - / - - / - - / - ms → zhs 25.43 / 83.82 28.23 / 89.38 30.85 / 90.27 31.57 / 88.14 34.01 / 89.68 25.17 / 90.12 23.82 / 87.07 25.35 / 90.31 27.32 / 87.44 27.48 / 91.73 32.73 / 91.67 32.85 / 88.90 28.15 / 88.79 - / - - / - - / - Table 14: Reference-based Chinese-centric evaluation results (spBLEU / XCOMET) of baseline models (Part I) on the FLORES+ benchmark. Direction Google Translate NLLB-54.5B Gemini 2.5 Pro Gemini 3 Pro GPT-5 MiLMMT-46-1B-v0.1 MiLMMT-46-1B-v1.0 MiLMMT-46-4B-v0.1 MiLMMT-46-4B-v1.0 MiLMMT-46-12B-v0.1 MiLMMT-46-12B-v1.0 zhs → ar 31.72 / 91.10 27.53 / 86.46 29.47 / 92.89 30.70 / 93.03 26.38 / 92.23 21.96 / 81.00 21.59 / 82.73 26.33 / 88.56 25.33 / 89.73 28.77 / 91.25 27.13 / 92.36 ar → zhs 37.23 / 87.26 21.25 / 70.04 34.46 / 89.09 35.09 / 89.37 32.71 / 89.05 26.16 / 80.95 26.13 / 82.10 30.85 / 85.73 30.42 / 87.01 32.55 / 88.06 32.00 / 89.16 zhs → az 21.35 / 79.59 19.24 / 70.95 22.51 / 86.13 22.90 / 86.77 20.43 / 84.90 16.86 / 66.61 16.48 / 69.88 19.86 / 79.83 18.91 / 81.08 21.21 / 84.04 20.46 / 84.95 az → zhs 30.41 / 79.10 19.32 / 64.19 27.77 / 82.79 28.12 / 82.86 25.42 / 83.07 21.06 / 70.93 20.83 / 73.74 25.17 / 79.29 24.55 / 80.79 26.56 / 81.08 25.76 / 82.83 zhs → bn 28.29 / 86.23 23.67 / 81.38 24.99 / 89.62 25.28 / 90.72 21.08 / 90.00 17.26 / 78.55 17.51 / 81.08 21.89 / 87.10 20.13 / 88.57 23.36 / 89.17 21.52 / 90.90 bn → zhs 34.34 / 82.48 20.01 / 63.34 31.52 / 85.02 32.57 / 85.09 30.31 / 85.45 23.20 / 72.94 23.47 / 75.43 28.46 / 80.47 27.97 / 82.44 30.24 / 83.32 29.62 / 85.00 zhs → bg 36.71 / 94.01 30.60 / 87.40 34.66 / 95.22 35.53 / 95.40 30.27 / 95.01 26.84 / 85.76 26.25 / 87.18 31.48 / 92.71 30.43 / 93.36 33.28 / 94.13 31.90 / 94.58 bg → zhs 37.56 / 86.59 20.07 / 62.52 34.37 / 88.34 35.17 / 88.18 32.55 / 87.96 28.37 / 81.95 28.18 / 83.75 32.38 / 86.17 31.79 / 87.91 33.90 / 87.30 33.13 / 89.11 zhs → ca 34.51 / 91.79 29.58 / 85.53 32.56 / 93.53 34.25 / 93.74 27.94 / 93.75 25.93 / 85.76 25.54 / 87.25 29.86 / 91.47 27.04 / 92.20 32.45 / 92.83 29.26 / 93.63 ca → zhs 38.27 / 85.84 23.32 / 67.61 34.95 / 88.07 35.72 / 87.80 33.38 / 88.15 29.71 / 81.70 29.46 / 83.30 33.30 / 86.21 33.11 / 87.89 34.29 / 87.57 33.77 / 89.37 zhs → cs 32.24 / 90.16 25.63 / 78.74 30.03 / 91.94 31.73 / 92.55 27.70 / 91.77 22.11 / 76.85 21.76 / 80.01 26.64 / 88.21 26.04 / 89.28 29.02 / 90.99 27.70 / 91.09 cs → zhs 37.49 / 83.38 19.60 / 57.22 34.24 / 85.74 34.53 / 85.46 32.31 / 86.04 28.46 / 78.64 28.46 / 80.74 32.24 / 83.61 32.10 / 85.52 33.89 / 85.03 33.21 / 87.11 zhs → zht 30.38 / 94.37 13.14 / 67.33 29.47 / 94.46 29.32 / 94.48 29.23 / 94.43 29.66 / 93.97 28.95 / 94.22 30.71 / 94.92 29.72 / 95.00 30.92 / 95.17 29.65 / 95.72 zht → zhs 28.49 / 91.58 24.10 / 82.35 29.61 / 91.96 29.96 / 92.11 29.56 / 92.15 31.77 / 92.93 32.15 / 93.39 33.74 / 93.91 33.84 / 94.66 34.44 / 94.20 29.34 / 94.67 zhs → da 34.41 / 95.48 28.32 / 90.49 32.07 / 96.75 33.17 / 97.09 29.61 / 97.00 24.93 / 89.35 24.37 / 90.58 29.37 / 94.76 28.43 / 95.27 31.85 / 96.17 30.41 / 96.39 da → zhs 38.86 / 90.50 19.51 / 62.21 35.39 / 91.64 35.93 / 91.39 33.70 / 91.39 30.25 / 86.40 29.91 / 87.30 34.18 / 89.96 33.94 / 91.35 34.76 / 90.93 34.00 / 92.20 zhs → de 33.71 / 97.85 27.15 / 93.92 31.92 / 98.08 32.77 / 98.04 29.21 / 97.89 23.95 / 93.60 23.30 / 94.71 28.45 / 96.90 27.55 / 97.29 30.64 / 97.68 29.22 / 97.92 de → zhs 37.98 / 91.31 21.48 / 71.34 34.80 / 92.50 34.91 / 92.33 32.66 / 92.20 29.24 / 87.59 29.04 / 89.08 32.77 / 90.75 31.89 / 92.09 34.04 / 92.23 33.15 / 93.00 zhs → el 31.35 / 91.24 26.00 / 84.84 31.29 / 92.90 31.95 / 93.14 28.72 / 92.75 22.27 / 79.61 22.55 / 81.50 27.07 / 89.97 26.35 / 90.52 29.29 / 92.08 28.39 / 92.55 el → zhs 35.41 / 82.39 20.76 / 64.25 32.70 / 84.88 33.23 / 84.44 30.77 / 84.92 26.16 / 76.52 25.63 / 78.50 29.71 / 82.27 29.18 / 84.70 32.05 / 84.16 31.11 / 86.04 zhs → en 39.59 / 97.18 36.13 / 95.04 38.26 / 97.28 38.12 / 97.38 34.82 / 97.52 32.44 / 95.14 30.86 / 95.53 35.40 / 96.59 33.76 / 96.79 36.60 / 97.21 34.99 / 97.36 en → zhs 44.05 / 93.84 26.61 / 76.72 39.42 / 94.26 40.26 / 94.43 37.73 / 94.19 35.03 / 90.32 34.23 / 91.75 38.03 / 93.21 36.85 / 94.54 38.98 / 94.11 37.30 / 95.32 zhs → tl 25.62 / 82.81 22.81 / 79.27 26.11 / 88.76 26.56 / 89.09 24.28 / 87.82 20.15 / 77.37 19.36 / 80.15 23.92 / 85.37 22.67 / 87.16 25.21 / 87.31 23.94 / 89.04 tl → zhs 38.06 / 79.55 21.61 / 61.36 34.81 / 81.94 35.80 / 81.73 32.34 / 81.99 27.68 / 73.83 27.17 / 75.28 32.15 / 79.13 31.34 / 81.46 33.62 / 80.86 32.45 / 82.63 zhs → fi 30.34 / 90.74 23.09 / 80.21 30.38 / 93.70 31.99 / 94.01 27.79 / 93.17 18.22 / 72.38 18.02 / 75.88 24.37 / 88.05 23.56 / 88.99 27.70 / 91.28 26.69 / 92.25 fi → zhs 35.43 / 84.37 18.72 / 57.59 33.03 / 86.40 33.75 / 86.81 31.33 / 86.97 26.03 / 78.18 26.02 / 79.88 31.06 / 84.61 30.38 / 85.84 32.53 / 86.08 31.46 / 87.96 zhs → fr 39.17 / 91.05 34.16 / 83.32 37.57 / 91.97 38.91 / 92.35 33.42 / 91.48 30.47 / 83.15 28.47 / 84.27 34.99 / 89.39 31.82 / 90.01 36.63 / 90.98 33.62 / 91.75 fr → zhs 38.27 / 86.91 21.19 / 63.19 34.52 / 89.06 34.94 / 88.71 32.83 / 88.93 29.68 / 83.12 29.66 / 85.07 32.67 / 87.05 31.94 / 88.43 34.00 / 88.04 33.47 / 90.30 zhs → he 34.40 / 90.22 27.04 / 84.83 30.38 / 92.79 32.47 / 92.88 27.35 / 92.58 18.70 / 75.10 19.36 / 77.41 26.33 / 88.70 25.01 / 89.71 29.13 / 91.37 27.85 / 92.54 he → zhs 38.02 / 88.74 19.77 / 65.10 35.10 / 90.40 35.51 / 90.49 33.13 / 89.97 26.47 / 80.89 26.31 / 82.34 31.83 / 87.45 31.59 / 88.62 33.39 / 89.13 32.72 / 90.40 zhs → hi 27.79 / 88.44 24.20 / 81.52 26.63 / 89.32 26.42 / 89.95 22.01 / 89.97 18.41 / 75.97 17.99 / 79.34 23.53 / 85.63 22.47 / 87.32 25.03 / 88.49 23.51 / 90.39 hi → zhs 36.62 / 86.57 21.12 / 69.64 33.67 / 88.38 34.04 / 88.02 31.37 / 87.73 25.66 / 77.85 25.67 / 80.20 30.13 / 84.63 29.83 / 86.09 31.82 / 86.55 31.09 / 88.31 zhs → hr 30.46 / 93.81 25.02 / 86.01 28.60 / 95.21 29.11 / 95.54 25.08 / 94.83 20.84 / 84.05 20.20 / 86.20 25.58 / 91.85 25.09 / 92.64 27.82 / 93.92 26.60 / 94.64 hr → zhs 36.56 / 88.00 18.81 / 60.71 33.02 / 89.75 34.15 / 89.57 31.91 / 89.56 27.85 / 82.57 27.88 / 84.48 31.82 / 87.66 31.88 / 89.29 32.94 / 89.07 32.49 / 90.53 zhs → hu 30.79 / 92.97 25.37 / 85.34 28.96 / 95.52 29.79 / 95.73 27.06 / 95.51 20.75 / 81.01 20.24 / 83.87 25.48 / 91.45 24.54 / 92.40 27.39 / 93.81 26.07 / 94.46 hu → zhs 36.72 / 87.07 19.59 / 62.74 32.53 / 89.37 33.56 / 89.32 31.28 / 89.15 26.98 / 80.49 26.98 / 82.69 30.93 / 87.28 30.74 / 88.75 32.84 / 88.69 32.15 / 90.39 zhs → id 34.25 / 94.54 29.55 / 89.89 30.52 / 95.55 31.42 / 95.54 28.09 / 95.67 26.19 / 90.95 24.36 / 91.83 29.65 / 94.38 27.44 / 95.08 30.05 / 95.09 27.26 / 95.71 id → zhs 38.33 / 90.92 21.38 / 69.69 34.49 / 92.06 35.66 / 91.83 33.24 / 91.46 29.15 / 86.61 29.39 / 87.49 32.97 / 89.88 32.66 / 90.81 34.00 / 91.37 33.92 / 92.57 zhs → it 30.56 / 93.67 25.77 / 86.63 30.78 / 94.94 30.93 / 94.79 27.18 / 94.56 22.91 / 87.64 22.29 / 88.40 26.49 / 92.46 25.05 / 92.87 29.10 / 94.20 26.87 / 94.59 it → zhs 35.88 / 89.11 19.33 / 62.66 32.01 / 90.83 32.74 / 90.97 30.30 / 91.24 27.13 / 85.77 27.09 / 87.37 30.67 / 89.16 30.19 / 90.85 31.69 / 90.41 30.47 / 91.71 zhs → ja 29.55 / 89.95 16.78 / 76.62 27.25 / 91.74 27.89 / 92.10 24.16 / 91.04 19.98 / 83.13 19.53 / 84.52 23.41 / 88.71 22.77 / 89.61 26.20 / 90.81 25.17 / 91.29 ja → zhs 34.07 / 82.96 19.64 / 62.43 28.99 / 86.22 29.61 / 85.80 28.14 / 86.22 24.47 / 79.53 24.61 / 81.16 28.02 / 84.29 28.25 / 85.64 29.85 / 85.56 29.57 / 87.24 zhs → k 28.77 / 75.87 22.99 / 66.92 25.47 / 82.04 26.50 / 83.29 23.09 / 81.11 18.27 / 62.21 18.22 / 64.78 22.69 / 76.25 22.21 / 77.30 25.26 / 80.52 23.91 / 81.35 k → zhs 35.71 / 77.97 19.31 / 56.48 32.85 / 81.00 33.22 / 80.92 30.81 / 81.08 25.05 / 70.34 24.67 / 73.01 29.35 / 77.57 29.05 / 79.09 30.79 / 79.40 30.65 / 81.38 zhs → km 22.14 / 70.91 17.55 / 67.55 23.49 / 78.21 24.33 / 79.32 20.28 / 75.91 17.17 / 63.13 17.32 / 66.31 21.48 / 75.36 21.66 / 76.89 22.93 / 77.66 23.06 / 79.58 km → zhs 33.35 / 73.35 21.33 / 61.75 31.67 / 77.95 32.44 / 77.75 27.29 / 77.41 22.73 / 66.56 21.60 / 69.31 27.82 / 73.99 26.30 / 76.21 28.96 / 76.29 27.23 / 77.86 zhs → ko 25.30 / 92.93 20.39 / 86.27 23.33 / 94.17 24.44 / 94.64 21.36 / 94.33 17.45 / 84.88 17.53 / 86.09 21.29 / 91.85 20.54 / 92.33 22.80 / 93.78 22.10 / 94.39 ko → zhs 34.47 / 83.78 21.31 / 70.27 30.52 / 86.77 31.66 / 86.45 30.01 / 86.76 25.57 / 79.09 25.58 / 80.42 29.13 / 83.80 28.75 / 85.34 30.52 / 85.73 30.57 / 87.13 zhs → lo 22.87 / 71.32 24.30 / 74.68 24.59 / 82.15 26.36 / 82.97 22.44 / 79.04 18.63 / 65.93 18.64 / 68.71 23.69 / 79.81 23.49 / 82.07 25.71 / 82.04 24.77 / 83.52 lo → zhs 34.36 / 78.31 23.44 / 69.34 32.51 / 81.55 33.35 / 81.56 29.15 / 81.05 23.38 / 70.29 23.29 / 71.69 28.42 / 77.38 27.94 / 79.29 30.43 / 79.53 30.09 / 81.48 zhs → my 20.67 / 64.57 15.79 / 60.31 21.92 / 77.87 22.65 / 78.89 16.00 / 72.64 10.99 / 53.13 13.16 / 56.01 18.72 / 73.18 18.61 / 74.57 21.33 / 77.21 19.96 / 78.88 my → zhs 30.84 / 63.59 16.77 / 46.61 29.17 / 68.97 30.19 / 69.82 25.85 / 68.15 18.62 / 52.56 18.35 / 55.47 24.01 / 62.85 23.58 / 64.69 26.19 / 65.54 25.56 / 67.61 zhs → nl 27.74 / 95.37 24.74 / 90.21 27.41 / 96.26 27.91 / 96.51 25.28 / 96.26 21.09 / 89.25 20.52 / 90.83 25.16 / 94.31 23.97 / 94.74 26.58 / 95.39 25.32 / 95.99 nl → zhs 33.58 / 89.39 16.61 / 61.24 30.53 / 90.46 30.92 / 90.52 28.50 / 89.70 25.91 / 84.99 25.90 / 86.64 28.83 / 88.81 28.67 / 90.20 29.97 / 90.20 29.53 / 91.45 zhs → nb 35.65 / 95.77 29.14 / 90.36 30.39 / 96.80 31.09 / 96.81 28.04 / 96.83 25.00 / 89.89 23.92 / 90.88 28.66 / 95.26 27.30 / 95.50 30.10 / 96.17 28.30 / 96.68 nb → zhs 41.74 / 89.96 18.51 / 56.28 37.11 / 90.99 38.16 / 90.89 35.40 / 90.73 31.14 / 84.64 30.87 / 86.32 34.99 / 88.69 34.74 / 89.93 36.50 / 89.84 35.86 / 91.51 zhs → fa 28.68 / 91.37 23.10 / 83.79 25.90 / 93.25 26.60 / 93.15 21.79 / 92.38 18.27 / 78.70 18.48 / 81.47 23.49 / 89.14 22.00 / 90.35 25.44 / 91.76 23.34 / 92.62 fa → zhs 35.67 / 87.52 21.89 / 70.96 32.86 / 89.85 33.86 / 89.49 31.60 / 89.68 24.56 / 79.81 24.77 / 82.18 30.62 / 86.92 30.23 / 88.18 31.90 / 88.65 31.53 / 89.77 zhs → pl 28.29 / 92.80 23.19 / 83.96 28.10 / 94.24 28.60 / 94.63 26.20 / 94.29 20.62 / 81.59 20.34 / 83.82 24.50 / 90.15 23.71 / 90.85 26.53 / 92.62 25.53 / 93.19 pl → zhs 33.75 / 84.48 17.37 / 58.21 30.44 / 86.31 31.08 / 86.47 28.73 / 87.16 25.64 / 80.76 25.79 / 82.23 29.00 / 84.93 28.80 / 86.59 30.29 / 86.16 30.08 / 88.28 zhs → pt 36.31 / 94.77 31.17 / 90.45 31.86 / 95.48 34.32 / 95.86 30.31 / 95.57 27.74 / 90.91 26.54 / 92.00 31.15 / 94.37 29.24 / 94.71 32.65 / 95.47 30.85 / 95.94 pt → zhs 38.41 / 91.12 20.54 / 66.68 35.11 / 92.44 35.43 / 92.23 33.24 / 91.73 29.59 / 87.22 29.57 / 88.33 33.50 / 90.53 33.16 / 92.04 34.41 / 91.53 33.67 / 92.84 zhs → ro 34.47 / 90.97 29.32 / 83.42 33.34 / 93.10 34.34 / 93.47 30.13 / 92.57 26.52 / 81.53 25.54 / 83.18 30.98 / 89.39 29.99 / 90.28 32.82 / 91.62 31.25 / 92.30 ro → zhs 38.34 / 84.90 22.97 / 65.44 35.09 / 87.72 35.88 / 87.43 32.77 / 87.62 29.80 / 81.03 29.49 / 82.72 33.36 / 85.60 32.69 / 87.39 34.47 / 86.73 33.37 / 88.48 zhs → ru 32.02 / 94.33 28.21 / 90.56 30.76 / 95.48 31.76 / 95.38 28.29 / 94.72 23.33 / 87.53 22.73 / 89.01 27.34 / 92.64 26.68 / 93.15 29.13 / 94.14 27.84 / 94.61 ru → zhs 36.32 / 90.50 21.22 / 74.88 32.03 / 92.08 33.43 / 92.15 31.25 / 92.07 27.76 / 86.84 27.70 / 88.03 31.09 / 90.37 30.61 / 91.43 32.87 / 91.65 31.83 / 92.72 zhs → sk 31.21 / 90.90 24.58 / 81.37 29.76 / 93.17 30.44 / 92.93 26.50 / 92.54 20.71 / 78.76 20.36 / 81.80 25.97 / 89.18 24.94 / 90.16 28.19 / 91.85 27.32 / 92.10 sk → zhs 36.91 / 84.83 19.35 / 58.54 33.84 / 86.85 34.18 / 86.92 31.92 / 87.12 27.83 / 79.52 27.87 / 81.56 31.72 / 84.94 31.10 / 86.60 33.32 / 86.46 32.73 / 88.25 zhs → sl 30.40 / 91.21 24.30 / 82.40 29.38 / 93.92 30.33 / 93.85 26.44 / 93.53 20.17 / 79.92 19.35 / 81.56 24.46 / 89.10 24.45 / 90.56 27.22 / 92.37 26.19 / 92.67 sl → zhs 35.82 / 85.09 18.68 / 58.85 32.68 / 87.81 33.75 / 87.42 31.01 / 87.59 26.79 / 80.39 26.83 / 82.17 30.88 / 85.43 30.60 / 86.94 32.11 / 87.35 31.70 / 88.63 zhs → es 26.08 / 94.70 23.73 / 90.24 26.59 / 95.47 26.62 / 95.47 25.34 / 95.26 21.62 / 90.87 21.47 / 91.82 24.75 / 94.17 24.70 / 94.74 25.45 / 94.96 25.33 / 95.57 es → zhs 33.97 / 90.46 18.16 / 63.15 30.76 / 92.12 31.98 / 92.12 29.18 / 91.94 26.54 / 86.13 25.98 / 87.40 29.68 / 90.08 28.87 / 91.49 30.43 / 91.11 29.58 / 92.67 zhs → sv 34.67 / 95.95 28.45 / 89.86 32.23 / 96.94 33.29 / 97.30 29.77 / 96.98 25.22 / 89.71 24.47 / 90.81 29.40 / 94.89 28.69 / 95.56 31.53 / 96.51 30.16 / 96.67 sv → zhs 38.68 / 90.66 20.55 / 66.10 35.24 / 91.52 35.70 / 91.83 33.48 / 91.68 29.62 / 85.63 29.39 / 86.98 33.32 / 90.22 33.09 / 91.57 34.87 / 91.50 34.06 / 92.52 zhs → ta 29.98 / 79.15 24.82 / 75.27 25.03 / 84.36 25.75 / 85.76 19.65 / 81.77 16.07 / 66.26 16.09 / 70.61 22.32 / 79.39 20.92 / 80.92 24.41 / 82.42 22.09 / 84.33 ta → zhs 33.00 / 76.72 20.43 / 62.99 30.82 / 80.20 31.70 / 79.90 29.07 / 80.57 20.92 / 65.44 20.60 / 67.84 26.09 / 74.95 25.60 / 77.32 28.48 / 77.67 27.93 / 80.23 zhs → th 38.61 / 89.01 30.20 / 80.13 37.77 / 92.96 38.66 / 93.04 35.94 / 92.46 29.26 / 82.05 29.41 / 84.76 34.92 / 90.56 33.08 / 91.34 37.14 / 92.05 34.68 / 93.33 th → zhs 35.07 / 87.53 20.07 / 69.89 32.59 / 89.75 33.06 / 89.80 30.64 / 89.71 25.50 / 82.64 25.18 / 83.86 29.38 / 87.71 29.10 / 89.07 31.63 / 89.61 30.70 / 90.55 zhs → tr 30.04 / 89.26 24.82 / 79.02 28.04 / 91.10 29.45 / 91.37 25.02 / 90.86 20.65 / 78.20 19.74 / 79.98 24.97 / 86.82 23.59 / 88.37 26.76 / 90.11 25.00 / 90.83 tr → zhs 37.21 / 85.99 20.27 / 61.39 33.15 / 88.10 34.01 / 87.59 31.40 / 88.13 26.79 / 78.68 26.61 / 80.53 30.86 / 85.15 30.94 / 86.87 32.62 / 86.73 32.08 / 88.63 zhs → ur 23.82 / 80.25 20.39 / 78.56 22.62 / 89.32 23.08 / 89.94 19.54 / 89.73 15.57 / 72.47 15.69 / 75.42 19.20 / 84.26 18.83 / 86.02 21.71 / 87.59 20.24 / 89.50 ur → zhs 34.48 / 80.66 21.79 / 67.90 32.14 / 84.65 32.80 / 83.80 30.16 / 84.06 22.70 / 70.28 22.17 / 72.45 27.86 / 79.09 27.75 / 81.10 30.49 / 82.04 29.73 / 83.79 zhs → uz 27.29 / 74.61 20.49 / 64.61 22.96 / 82.19 24.12 / 82.12 20.92 / 80.07 16.20 / 61.46 14.95 / 64.78 19.60 / 74.42 17.96 / 76.36 21.89 / 79.45 19.57 / 80.95 uz → zhs 35.88 / 73.19 21.88 / 61.23 32.38 / 77.02 33.03 / 76.83 30.33 / 77.61 24.65 / 65.38 24.50 / 67.58 29.12 / 72.70 28.84 / 75.03 31.30 / 75.80 30.88 / 77.40 zhs → vi 35.80 / 93.03 31.75 / 88.98 32.61 / 93.92 32.95 / 94.18 30.89 / 94.16 29.52 / 89.63 28.63 / 90.79 32.12 / 92.90 30.80 / 93.57 32.79 / 93.66 30.95 / 94.45 vi → zhs 36.04 / 90.47 21.79 / 74.70 33.00 / 91.80 33.94 / 91.62 31.06 / 91.50 27.49 / 85.90 27.57 / 87.47 31.09 / 89.80 30.88 / 90.89 32.56 / 91.01 32.04 / 92.26 zhs → yue 35.08 / 92.38 22.02 / 83.28 35.66 / 92.12 33.00 / 90.55 36.48 / 93.06 31.67 / 92.29 33.25 / 92.29 34.76 / 93.59 34.22 / 93.74 34.97 / 94.01 35.01 / 94.11 yue → zhs 34.21 / 92.11 27.95 / 85.92 36.46 / 93.93 37.38 / 94.23 37.96 / 94.47 35.10 / 92.45 35.39 / 93.11 37.71 / 94.18 37.23 / 94.92 39.00 / 94.69 38.82 / 95.26 zhs → ms 30.43 / 89.91 26.33 / 88.52 26.95 / 93.94 27.64 / 93.98 24.65 / 94.18 22.77 / 89.51 22.16 / 90.48 25.88 / 93.03 24.37 / 93.61 27.00 / 93.71 24.78 / 94.82 ms → zhs 37.54 / 89.48 23.16 / 73.35 33.84 / 90.82 34.89 / 90.85 32.31 / 90.41 28.14 / 84.56 28.25 / 85.95 32.32 / 88.89 32.39 / 90.04 33.38 / 90.21 33.28 / 91.52 Table 15: Reference-based Chinese-centric evaluation results (spBLEU / XCOMET) of baseline models (Part I) and MiLMMT models on the FLORES+ benchmark. Appendix E Full On-Policy Distillation Results OPD Training Details. We adopt optimization settings similar to those used for GRPO in our OPD experiments, including a learning rate of 1×10−61× 10^-6, a prompt batch size of 128128, a PPO mini-batch size of 128128, three training epochs, and maximum prompt and response lengths of 40964096 tokens each. For RL+OPD, we use the same task reward and group size G=8G=8 as in GRPO. Pure OPD does not use the task reward and instead samples one student trajectory per prompt. We use the fixed MiLMMT-46-12B-v1.0 model as the teacher and the k1k_1 estimator for PG-OPD. For the GKD control, we use a top-128128 approximation to the forward KL divergence. Full Results. Table 16 reports the complete OPD and RL+OPD results underlying Section 5 and Table 2, as 4646-language macro-averages. All students are distilled from the fixed MiLMMT-46-12B-v1.0 teacher. RL+OPD adds the QE-based task reward with distillation weight λ; OPD rows use only the distillation loss. Unless noted, the student is initialized from v0.1; init. v1.0 rows are initialized from the interpolated checkpoint instead. WMT24++ cells report reference-free XCOMET / COMETKiwi; each FLORES+ direction group reports spBLEU / reference-based XCOMET / reference-free XCOMET / COMETKiwi. Model WMT24++ FLORES+ en → x en → x x → en zh → x x → zh 11B students RL+OPD λ=1λ=1 78.25/75.35 33.87/87.79/90.71/86.56 39.65/91.17/90.65/88.27 21.64/82.24/82.76/77.77 26.77/81.48/67.18/76.93 RL+OPD λ=0.5λ=0.5 78.58/75.69 33.76/87.90/90.90/86.72 39.41/91.24/90.75/88.30 21.63/82.43/82.97/78.00 26.70/81.57/67.50/77.10 RL+OPD λ=0.1λ=0.1 79.37/76.41 33.32/88.38/91.39/87.19 38.56/91.44/91.02/88.36 21.34/82.93/83.66/78.61 26.40/81.89/68.14/77.43 RL+OPD λ=0.05λ=0.05 79.83/76.68 32.66/88.49/91.56/87.29 38.02/91.47/91.11/88.37 21.04/83.00/83.72/78.80 26.04/82.09/68.39/77.84 RL+OPD λ=0.01λ=0.01 80.29/77.01 31.11/88.53/91.74/87.29 37.33/91.45/91.14/88.33 20.47/83.12/83.95/78.97 24.89/82.10/68.74/77.78 RL+OPD λ=0.001λ=0.001 80.42/77.08 29.65/88.37/91.66/87.29 36.30/91.46/91.15/88.24 19.70/82.82/83.82/79.14 24.12/82.05/68.51/77.79 RL+OPD λ=10−4λ=10^-4 80.26/76.92 30.03/88.32/91.62/87.19 36.65/91.45/91.11/88.27 19.89/82.82/83.93/79.08 24.18/82.04/69.08/77.97 OPD (PG) 77.67/74.66 33.95/87.19/90.21/86.03 39.78/90.97/90.42/88.19 21.44/81.63/82.13/77.17 26.70/81.07/66.70/76.58 OPD (GKD) 77.42/74.37 32.90/87.12/90.11/85.91 39.54/90.96/90.44/88.15 20.83/81.32/81.77/76.86 26.35/81.02/66.56/76.40 OPD (PG, init. v1.0) 77.86/74.93 33.87/87.45/90.43/86.29 39.58/91.06/90.54/88.23 21.43/81.93/82.40/77.42 26.69/81.23/66.87/76.68 44B students RL+OPD λ=1λ=1 85.85/82.23 37.09/93.34/95.15/91.12 42.70/93.98/92.78/89.41 25.44/89.92/88.56/82.36 30.68/86.67/70.43/79.99 RL+OPD λ=0.5λ=0.5 86.03/82.45 36.81/93.44/95.27/91.27 42.43/94.03/92.86/89.42 25.22/89.96/88.60/82.51 30.43/86.89/70.59/80.17 RL+OPD λ=0.1λ=0.1 86.67/82.95 35.40/93.72/95.57/91.57 41.59/94.11/93.00/89.42 24.55/90.13/88.93/82.85 29.67/87.27/71.50/80.57 RL+OPD λ=0.05λ=0.05 87.03/83.17 34.25/93.74/95.62/91.66 41.22/94.10/93.03/89.41 24.01/90.09/88.97/82.95 29.18/87.30/71.36/80.55 RL+OPD λ=0.01λ=0.01 87.62/83.46 31.76/93.89/95.88/91.77 41.06/94.16/92.99/89.39 23.38/90.19/88.96/83.28 28.19/87.61/71.41/80.66 RL+OPD λ=0.001λ=0.001 87.68/83.57 30.35/93.87/95.88/91.80 40.64/94.13/92.95/89.38 22.70/90.16/89.03/83.45 27.12/87.71/71.98/80.94 RL+OPD λ=10−4λ=10^-4 87.74/83.45 30.41/93.85/95.83/91.74 40.27/94.14/92.99/89.36 22.63/90.06/89.01/83.52 26.63/87.63/71.56/80.67 OPD (PG) 85.42/81.86 37.56/93.15/94.97/90.94 42.88/93.91/92.68/89.37 25.46/89.74/88.36/82.13 30.80/86.47/70.01/79.79 OPD (GKD) 85.62/81.96 36.97/93.21/95.01/91.01 42.68/93.94/92.76/89.38 25.28/89.76/88.36/82.25 30.71/86.58/70.08/79.88 OPD (PG, init. v1.0) 85.37/81.85 37.53/93.14/94.97/90.97 42.78/93.89/92.71/89.40 25.47/89.69/88.32/82.09 30.85/86.45/70.08/79.76 Table 16: Full on-policy distillation results as 4646-language. All students are distilled from the fixed MiLMMT-46-12B-v1.0 teacher. WMT24++ cells report reference-free XCOMET / COMETKiwi; each FLORES+ cell reports spBLEU / reference-based XCOMET / reference-free XCOMET / COMETKiwi.

Instant research discovery