Paper deep dive
The Rise of Verbal Tics in Large Language Models: A Systematic Analysis Across Frontier Models
Shuai Wu, Xue Li, Yanna Feng, Yufang Li, Zhijun Wang, Ran Wang
Intelligence
Status: succeeded | Model: Gemma-4-26B-A4B | Prompt: intel-v1 | Confidence: 98%
Last extracted: 4/26/2026, 10:55:50 PM
Summary
This technical report presents a systematic analysis of 'verbal tics'ârepetitive, formulaic linguistic patternsâin eight frontier Large Language Models (LLMs). The researchers introduce the Verbal Tic Index (VTI) to quantify these phenomena, which include sycophantic openers, pseudo-empathetic affirmations, and overused vocabulary. The study reveals that models like Gemini 3.1 Pro exhibit higher VTI scores compared to models like DeepSeek V3.2. Key findings include the 'alignment tax,' where alignment techniques like RLHF contribute to increased sycophancy and decreased perceived naturalness, and a temporal accumulation effect where verbal tics increase during multi-turn conversations. The analysis covers both English and Chinese, demonstrating task-dependent and cross-lingual variations in tic prevalence.
Entities (9)
Relation Signals (4)
Gemini 3.1 Pro â hasvtiscore â 0.590
confidence 100% ¡ Gemini 3.1 Pro exhibits the highest VTI (0.590)
DeepSeek-V3.2 â hasvtiscore â 0.295
confidence 100% ¡ DeepSeek V3.2 achieves the lowest (0.295)
Sycophancy â inverselycorrelatedwith â Naturalness
confidence 100% ¡ confirms a strong inverse relationship between sycophancy and perceived naturalness (r = -0.87, p < 0.001)
RLHF â contributesto â Sycophancy
confidence 90% ¡ As Large Language Models (LLMs) continue to evolve through alignment techniques such as Reinforcement Learning from Human Feedback (RLHF)... a growing and increasingly conspicuous phenomenon has emerged: the proliferation of verbal tics
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:As Large Language Models (LLMs) continue to evolve through alignment techniques such as Reinforcement Learning from Human Feedback (RLHF) and Constitutional AI, a growing and increasingly conspicuous phenomenon has emerged: the proliferation of verbal tics -- repetitive, formulaic linguistic patterns that pervade model outputs. These range from sycophantic openers ("That's a great question!", "Awesome!") to pseudo-empathetic affirmations ("I completely understand your concern", "I'm right here to catch you") and overused vocabulary ("delve", "tapestry", "nuanced"). In this paper, we present a systematic analysis of the verbal tic phenomenon across eight state-of-the-art LLMs: GPT-5.4, Claude Opus 4.7, Gemini 3.1 Pro, Grok 4.2, Doubao-Seed-2.0-pro, Kimi K2.5, DeepSeek V3.2, and MiMo-V2-Pro. Utilizing a custom evaluation framework for standardized API-based evaluation, we assess 10,000 prompts across 10 task categories in both English and Chinese, yielding 160,000 model responses. We introduce the Verbal Tic Index (VTI), a composite metric quantifying tic prevalence, and analyze its correlation with sycophancy, lexical diversity, and human-perceived naturalness. Our findings reveal significant inter-model variation: Gemini 3.1 Pro exhibits the highest VTI (0.590), while DeepSeek V3.2 achieves the lowest (0.295). We further demonstrate that verbal tics accumulate over multi-turn conversations, are amplified in subjective tasks, and show distinct cross-lingual patterns. Human evaluation (N = 120) confirms a strong inverse relationship between sycophancy and perceived naturalness (r = -0.87, p < 0.001). These results underscore the "alignment tax" of current training paradigms and highlight the urgent need for more authentic human-AI interaction frameworks.
Tags
Links
- Source: https://arxiv.org/abs/2604.19139v1
- Canonical: https://arxiv.org/abs/2604.19139v1
Trouble viewing inline? Open PDF directly â
Full Text
44,277 characters extracted from source content.
Expand or collapse full text
THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELS: A SYSTEMATIC ANALYSIS ACROSS FRONTIER MODELS TECHNICAL REPORT Shuai Wu * M.S. Candidate, Lead Researcher Xue Li M.S., Research Assistant Yanna Feng Ph.D. Candidate, Academic Advisor Yufang Li Ph.D., Academic Advisor Zhijun Wang Ph.D., Research Consultant Ran Wang B.S., Research Assistant April 2026 ABSTRACT As Large Language Models (LLMs) continue to evolve through alignment techniques such as Rein- forcement Learning from Human Feedback (RLHF) and Constitutional AI, a growing and increasingly conspicuous phenomenon has emerged: the proliferation of verbal ticsârepetitive, formulaic lin- guistic patterns that pervade model outputs. These range from sycophantic openers (âThatâs a great question!â, âAwesome!â) to pseudo-empathetic affirmations (âI completely understand your con- cernâ, âIâm right here to catch youâ) and overused vocabulary (âdelveâ, âtapestryâ, ânuancedâ). In this paper, we present a systematic analysis of the verbal tic phenomenon across eight state-of-the-art LLMs: GPT-5.4, Claude Opus 4.7, Gemini 3.1 Pro, Grok 4.2, Doubao-Seed-2.0-pro, Kimi K2.5, DeepSeek V3.2, and MiMo-V2-Pro. Utilizing a custom evaluation framework for standardized API-based evaluation, we assess 10,000 prompts across 10 task categories in both English and Chinese, yielding 160,000 model responses. We introduce the Verbal Tic Index (VTI), a composite metric quantifying tic prevalence, and analyze its correlation with sycophancy, lexical diversity, and human-perceived naturalness. Our findings reveal significant inter-model variation: Gemini 3.1 Pro exhibits the highest VTI (0.590), while DeepSeek V3.2 achieves the lowest (0.295). We further demonstrate that verbal tics accumulate over multi-turn conversations, are amplified in subjective tasks, and show distinct cross-lingual patterns. Human evaluation (N = 120) confirms a strong inverse relationship between sycophancy and perceived naturalness (r =â0.87,p < 0.001). These results underscore the âalignment taxâ of current training paradigms and highlight the urgent need for more authentic human-AI interaction frameworks. Keywords Large Language Models¡Verbal Tics¡Sycophancy¡RLHF¡Alignment Tax¡Lexical Diversity¡ Catchphrases¡ Pseudo-Empathy 1 Introduction The rapid advancement of Large Language Models (LLMs) has fundamentally transformed the landscape of human- computer interaction. Models such as GPT-5.4 (OpenAI, 2026), Claude Opus 4.7 (Anthropic, 2026), Gemini 3.1 Pro (Google DeepMind, 2026), and their contemporaries now serve as conversational assistants, creative collaborators, and knowledge workers across billions of interactions daily. A critical enabler of this success has been alignment trainingâthe process of fine-tuning models to be helpful, harmless, and honest through techniques like Reinforcement Learning from Human Feedback (RLHF) (Ouyang et al., 2022) and Constitutional AI (Bai et al., 2022). However, as these alignment techniques have matured and been applied at scale, a distinct and increasingly conspicuous linguistic artifact has emerged: the verbal tic. We define a verbal tic as a repetitive, formulaic expression or phrase that â Corresponding author. Email: noah.wu@tuta.io arXiv:2604.19139v1 [cs.CL] 21 Apr 2026 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT appears with disproportionate frequency in a modelâs output, independent of the specific conversational context. These tics manifest in several forms: â˘Sycophantic openers: Exaggerated praise or validation of the userâs input (e.g., âThatâs a great question!â, âExcellent observation!â, âYour insight is incredibly sharp!â). â˘Pseudo-empathetic affirmations: Formulaic emotional understanding that often feels hollow (e.g., âI completely understand your concernâ, âIâm right here, not hiding, not dodging, ready to catch youâ). â˘Hedging phrases: Defensive language designed to soften assertions (e.g., âItâs important to note that...â, âI have to be honest...â). â˘Overused vocabulary: Specific words that appear with statistically anomalous frequency (e.g., âdelveâ, âtapestryâ, ânuancedâ, âmultifacetedâ). â˘Filler transitions: Unnecessary connective phrases that pad responses (e.g., âFurthermoreâ, âMoreoverâ, âLet me walk you through this step by stepâ). This phenomenon has been widely discussed in both academic literature and public discourse. Sharma et al. (2023) provided early evidence that RLHF-trained models exhibit systematic sycophancy across multiple evaluation paradigms. More recently, Cheng et al. (2026) demonstrated in Science that sycophantic AI responses decrease prosocial intentions and promote dependence in users (N = 2405). The Stanford AI Index Report of 2026 (Stanford HAI, 2026) further highlighted declining model transparency, raising concerns about the mechanisms driving these behaviors. In this paper, we present a systematic, cross-model, cross-lingual analysis of verbal tics in frontier LLMs. Our contributions include: 1. A systematic taxonomy of verbal tics across English and Chinese, with fine-grained categorization. 2. The Verbal Tic Index (VTI), a composite metric for standardized measurement. 3.Large-scale evaluation of 8 frontier models across 10 task types, 10 prompt complexity levels, and 20-turn conversa- tions. 4. Cross-lingual analysis revealing distinct tic patterns between English and Chinese. 5. Human evaluation (N = 120) correlating VTI with perceived naturalness, helpfulness, and trust. 6. Embedding-space analysis of tic phrases using t-SNE visualization. 2 Related Work 2.1 Sycophancy in Language Models The study of sycophancy in LLMs has gained significant attention since the seminal work of Sharma et al. (2023), who identified consistent patterns of sycophantic behavior across five state-of-the-art AI assistants. Their analysis revealed that models trained with RLHF are particularly susceptible to conforming to user beliefs, even when those beliefs are factually incorrect. Carro (2024) further explored the impact of sycophantic tendencies on user trust, finding a complex relationship between perceived agreeableness and long-term credibility. Batzner et al. (2025) reviewed methodological challenges in measuring LLM sycophancy, identifying five core operationalizations and highlighting the inherently human nature of the phenomenon. Most recently, Cheng et al. (2026) published a landmark study in Science demonstrating that sycophantic AI responses not only affect user perception but actively promote dependence and reduce prosocial intentions, with implications for decision-making across multiple domains. 2.2 Repetition and Verbal Patterns in LLMs The ârepeat curseâ in LLMsâthe tendency for models to generate repetitive textâhas been studied from multiple perspectives. Yao et al. (2025) investigated the root causes of repetition through mechanistic interpretability, locating specific layers involved in repetitive generation via Sparse Autoencoder-based activation manipulation. Xu et al. (2022) proposed the DITTO framework, a training-time penalty for pseudo-repetition at NeurIPS 2022, demonstrating that repetition can be mitigated without sacrificing generation quality. 2 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT 2.3 AI Detection and Linguistic Fingerprints The proliferation of verbal tics has also driven research in AI-generated text detection. Detectors leverage metrics such as perplexity and burstiness to identify the statistical predictability characteristic of LLM outputs (Mitchell et al., 2023). The observation that specific phrases (e.g., âdelveâ, âtapestryâ) serve as reliable indicators of AI authorship has been widely documented in both academic and popular literature. 2.4 Medical and Domain-Specific Sycophancy Kim et al. (2026) evaluated sycophantic behavior in ten LLMs across multi-turn medical conversations using an escalatory pushback framework, finding that all models are more easily persuaded to change their answers on clear multiple-choice questions than on ambiguous diagnostic cases. Their work highlights critical vulnerabilities in deploying LLMs for clinical decision support. 3 Methodology 3.1 Evaluated Models Our study evaluates eight state-of-the-art LLMs representing diverse architectural paradigms, training methodologies, and organizational origins. Table 1 summarizes the key characteristics of each model. Table 1: Overview of evaluated models. All models were accessed via a unified evaluation framework using their respective API endpoints. ModelDeveloperAccessNotes GPT-5.4OpenAIAPILatest GPT series Claude Opus 4.7AnthropicAPIConstitutional AI Gemini 3.1 ProGoogle DeepMindAPIMultimodal capable Grok 4.2xAIAPIReal-time knowledge Doubao-Seed-2.0-proByteDanceAPIChinese-optimized Kimi K2.5Moonshot AIAPILong-context specialist DeepSeek V3.2DeepSeekAPIMoE architecture MiMo-V2-ProXiaomiAPIReasoning-focused 3.2 Experimental Platform All experiments were conducted through a custom evaluation framework 2 that provides unified API access to all evaluated models. This framework supports standardized request formatting, response parsing, and logging across providers, ensuring experimental consistency. The architecture routes requests through provider-specific API adapters while maintaining a common interface for prompt injection, parameter control, and response collection. 3.3 Dataset Construction We constructed a diverse evaluation dataset of 10,000 prompts spanning 10 task categories, designed to elicit a wide range of conversational behaviors: Each prompt was presented in both English and Chinese, yielding 20,000 total interactions per model and 160,000 total responses across all eight models. All API calls were made between March 1â15, 2026, using the model versions available at that time. 3.4 Verbal Tic Detection Pipeline We developed an automated verbal tic detection pipeline consisting of three stages: 1. Lexical Matching: Pattern-based detection of known tic phrases using a curated dictionary of 200+ English and 150+ Chinese verbal tics (see Appendix A). Each phrase is matched with context-aware rules to reduce false positivesâfor 2 https://github.com/Noah-Wu66/Vectaix-AI 3 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT Table 2: Task categories and their descriptions. Task CategoryDescriptionPrompts Creative WritingFiction, poetry, storytelling1,000 Code GenerationProgramming tasks across languages1,000 Math ReasoningMathematical problem-solving1,000 Casual ChatOpen-ended conversation1,000 Academic Q&AScholarly questions and explanations1,000 Emotional SupportEmpathetic dialogue and counseling1,000 Debate/ArgumentPersuasive and adversarial dialogue1,000 SummarizationText compression and synthesis1,000 TranslationCross-lingual translation tasks1,000 Role-PlayingCharacter-based interaction1,000 example, âAbsolutely!â is only flagged as a sycophantic opener when it appears at the beginning of a response, not when used as an adverb within a sentence. 2.Statistical Analysis: Identification of statistically over-represented n-grams (n â 1, 2, 3, 4) using TF-IDF weighting against a human-written reference corpus (a balanced sample of 50,000 sentences from Wikipedia and Reddit). 3.Semantic Clustering: Embedding-based grouping of semantically similar tic phrases using theall-MiniLM-L6-v2 sentence transformer, enabling detection of paraphrased tics. Phrases with cosine similarity> 0.85are merged into the same tic cluster. When a phrase matches multiple categories (e.g., âAbsolutely!â as both a sycophantic opener and an emphatic affirmation), we assign it to the category with the highest contextual probability based on its position and surrounding tokens. This priority rule prevents double-counting across categories. 3.5 The Verbal Tic Index (VTI) We define the Verbal Tic Index (VTI) as a composite metric: VTI = ι¡ TicRate + β¡ (1â TTR norm ) + γ¡ SycScore + δ¡ RepRate(1) where: ⢠TicRateâ [0, 1]: Proportion of responses containing at least one detected verbal tic. ⢠TTR norm : Length-normalized Type-Token Ratio, computed over a fixed sliding window of 200 tokens (MATTR) to mitigate length sensitivity. English tokenization uses spaCy; Chinese tokenization uses jieba. ⢠SycScoreâ [0, 1]: Sycophancy score based on the proportion of sycophantic openers and pseudo-empathetic phrases. ⢠RepRateâ [0, 1]: Repetition rate of unique phrases across responses. ⢠ι = 0.3,β = 0.2,Îł = 0.3,δ = 0.2: Weights determined through a grid search over0.1, 0.2, 0.3, 0.4to maximize rank correlation with human judgments on a held-out validation set of 500 annotated responses. 3.6 Human Evaluation Protocol We recruited 120 human evaluators (60 English-speaking, 60 Chinese-speaking) from a university participant pool. Each evaluator assessed 50 randomly sampled responses on a 5-point Likert scale across six dimensions: Naturalness, Helpfulness, Sycophancy Perception, Trust, Annoyance, and Repetitiveness. Likert anchors ranged from 1 (âNot at allâ) to 5 (âExtremelyâ). Evaluators were blinded to model identity. Response-evaluator assignment was randomized, yielding 6,000 total annotations. Inter-annotator agreement was measured using KrippendorffâsÎą = 0.72, indicating substantial agreement. 4 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT Table 3: Complete Verbal Tic Index (VTI) scores and component metrics for all evaluated models. ModelVTI (EN)VTI (ZH)VTI (All)Syc. IndexDiv. IndexNat. Index GPT-5.40.4230.3980.4110.4560.5670.589 Claude Opus 4.70.2890.3450.3170.3120.6780.734 Gemini 3.1 Pro0.5670.6120.5900.6340.4890.445 Grok 4.20.3450.3120.3290.3780.6120.634 Doubao-Seed-2.0-pro0.4450.4890.4670.5230.5340.556 Kimi K2.50.3890.4230.4060.4670.5780.601 DeepSeek V3.20.3120.2780.2950.2980.6450.689 MiMo-V2-Pro0.3780.4010.3900.4230.5120.523 4 Results 4.1 Overall Verbal Tic Index Figure 1 presents the VTI scores across all models in English, Chinese, and overall. The results reveal significant inter-model variation, with VTI scores ranging from 0.295 (DeepSeek V3.2) to 0.590 (Gemini 3.1 Pro). GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Verbal Tic Index (VTI) 0.411 0.317 0.590 0.329 0.467 0.406 0.295 0.390 Verbal Tic Index Comparison Across Models VTI (English) VTI (Chinese) VTI (Overall) Figure 1: Verbal Tic Index (VTI) comparison across models. Lower scores indicate fewer verbal tics and more natural language use. Table 3 provides the complete VTI breakdown with component scores. The Diversity Index and Naturalness Index are derived from the human evaluation and lexical diversity metrics, respectively, where higher values indicate better performance (more diverse and more natural). 4.2 Multi-Dimensional VTI Analysis Figure 2 provides a radar chart visualization of the multi-dimensional VTI profile. To ensure a consistent interpretation where larger polygon areas correspond to more problematic verbal tic behavior, the Diversity Index and Naturalness Index axes are inverted (plotted as 1â value), so that all six axes follow the convention of âhigher is worse.â 5 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT VTI Overall Sycophancy Index Diversity Index Naturalness Index VTI (English) VTI (Chinese) 0.2 0.4 0.6 Multi-Dimensional Verbal Tic Index (VTI) Comparison GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 Figure 2: Multi-dimensional VTI profile radar chart. All axes are oriented so that higher values indicate more problematic behavior (Diversity and Naturalness axes are inverted). Larger polygon areas thus correspond to more pronounced verbal tic profiles. Gemini 3.1 Pro shows the most expansive profile, while Claude Opus 4.7 and DeepSeek V3.2 maintain compact, low-tic profiles. 4.3 English Verbal Tic Frequency Distribution Figure 3 presents the frequency of English verbal tic categories per 1,000 responses. Emphatic Affirmations and Overused Vocabulary are the most prevalent categories across models, with Gemini 3.1 Pro showing particularly high rates of Emphatic Affirmations (523 per 1,000 responses). 4.4 Chinese Verbal Tic Frequency Distribution The Chinese verbal tic landscape reveals distinct patterns (Figure 4). Sycophantic Openers dominate in Gemini 3.1 Pro (567/1000) and Doubao-Seed-2.0-pro (423/1000), while Pseudo-Empathy is most pronounced in Claude Opus 4.7 (456/1000) and GPT-5.4 (345/1000). 6 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 Model Sycophantic Openers Hedging Phrases Filler Transitions Emphatic Affirmations Pseudo-Empathy Overused Vocabulary Verbal Tic Category 312145487278356298189267 267312234198287245278312 345189398256312278234289 289167523312378334212298 234389312178267234198256 378234289312198267345223 English Verbal Tic Frequency by Category and Model 150 200 250 300 350 400 450 500 Frequency per 1,000 Responses Figure 3: English verbal tic frequency heatmap. Values represent occurrences per 1,000 responses. Darker cells indicate higher frequencies. GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 Model Sycophantic Openers Pseudo-Empathy False Modesty Excessive Emphasis Formulaic Transitions Verbal Tic Category 298134567245423378178312 345456289167312256198278 189478178234198189267223 267123534289398345167298 312267345234289278256267 Chinese Verbal Tic Frequency by Category and Model 150 200 250 300 350 400 450 500 550 Frequency per 1,000 Responses Figure 4: Chinese verbal tic frequency heatmap. The distribution reveals that Chinese-language tics are more concen- trated in sycophantic and pseudo-empathetic categories. 7 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT 4.5 Top Verbal Tic Phrases Figure 5 and Figure 6 present the most frequently occurring verbal tic phrases in English and Chinese, aggregated across all models. 0100200300400500600 Total Frequency Across All Models tapestry holistic multifaceted I'd be happy to help leverage landscape Great observation! nuanced Let me break this down robust Furthermore, delve It's important to note That's a great question! Absolutely! 283 338 338 402 426 448 449 470 490 514 578 590 611 617 654 Top English Verbal Tic Phrases (Aggregated) Figure 5: Top 15 English verbal tic phrases by total frequency across all models. âAbsolutely!â and âThatâs a great question!â lead the rankings, followed closely by âItâs important to noteâ and âdelveâ. 0100200300400500600 Total Frequency Across All Models An academic gold mine Caliber of a top-journal author This is my most honest answer so far Not because you're wrong but too right This question makes me uneasy First I must strongly congratulate you Ready to catch you I have to be honest You've hit the core of the issue This line of thinking is brilliant I have to say this very seriously Your insight is incredibly sharp Let me walk you through step by step Awesome!/Amazing! This is a really great question 184 204 211 224 225 259 277 317 338 346 349 368 371 546 604 Top Chinese Verbal Tic Phrases (Translated, Aggregated) Figure 6: Top 15 Chinese verbal tic phrases by total frequency. âThis is a really great questionâ and âAwe- some!/Amazing!â are the most prevalent catchphrases. 8 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT 4.6 Task-Dependent Verbal Tic Rates The prevalence of verbal tics varies dramatically across task types (Figure 7). Emotional Support tasks elicit the highest tic rates (mean = 0.55 across models), followed by Role-Playing (0.49) and Debate/Argument (0.39). Conversely, Translation (0.09) and Code Generation (0.13) tasks produce the fewest tics, likely because these tasks demand precise, structured outputs with less room for conversational filler. Creative Writing Code Generation Math Reasoning Casual Chat Academic Q&A Emotional Support Debate/Argument Summarization Translation Role-Playing Task Type 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Verbal Tic Rate Verbal Tic Rate by Task Type and Model GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 Figure 7: Verbal tic rate by task type and model. Subjective, conversational tasks consistently elicit higher rates of formulaic responses across all models. 4.7 Temporal Dynamics: The Accumulation Effect A notable finding of our study is the temporal accumulation of verbal tics over multi-turn conversations (Figure 8). Across all models, the verbal tic rate shows a clear overall upward trend from Turn 1 to Turn 20, with an average increase of approximately 110% from the first to the last turn. This pattern is consistent with the ârepeat curseâ identified by Yao et al. (2025) and the multi-turn sycophancy escalation observed by Kim et al. (2026), suggesting that models progressively fall into repetitive linguistic loops as context length grows. 9 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT 2.55.07.510.012.515.017.520.0 Conversation Turn 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Verbal Tic Rate Verbal Tic Rate Across Conversation Turns GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 Figure 8: Verbal tic rate across 20 conversation turns. All models show increasing reliance on tics as conversations progress, with GPT-5.4 and MiMo-V2-Pro showing the steepest increases. 4.8 Sycophancy Analysis by Prompt Type The sycophancy score varies significantly based on prompt type (Figure 9). Emotional Appeal prompts trigger the highest sycophancy scores (mean = 0.68), followed by Praise Seeking (0.61) and Self-Deprecation (0.57). Neutral queries and technical questions produce the lowest sycophancy scores (mean = 0.23 and 0.20, respectively). GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 Model Neutral Query Leading Question Emotional Appeal Authority Claim Self-Deprecation Praise Seeking Controversial Topic Technical Question Prompt Type 0.2290.3090.2650.1890.2190.2590.1180.215 0.3130.4440.2980.4650.2340.2160.4250.391 0.7780.6430.8340.5050.6720.6310.5090.830 0.2950.2610.4580.2410.3660.4190.4050.335 0.5540.4550.5940.4160.4610.5760.7290.736 0.4770.5380.6640.4640.6020.7500.6510.767 0.2090.3920.3470.4990.3390.4310.3360.269 0.2930.2430.1500.2410.2230.1330.2290.114 Sycophancy Score by Prompt Type and Model 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Sycophancy Score Figure 9: Sycophancy score heatmap by prompt type and model. Emotionally charged prompts consistently elicit higher sycophancy across all models. 10 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT 4.9 Sycophancy vs. Naturalness: The Alignment Tax Figure 10 reveals a strong inverse correlation between a modelâs Sycophancy Index and its human-rated Naturalness score (r =â0.87,p < 0.001). Models that rely heavily on sycophantic openers and pseudo-empathy are consistently perceived as less natural and more âroboticâ by human evaluators. Notably, the bubble sizes (representing Helpfulness scores) show that sycophancy does not necessarily improve perceived helpfulnessâClaude Opus 4.7 achieves the highest Helpfulness score (4.45/5) while maintaining the lowest Sycophancy Index (0.312). 0.30.40.50.60.7 Sycophancy Index 2.6 2.8 3.0 3.2 3.4 3.6 3.8 4.0 Human-Rated Naturalness (1-5) GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 Sycophancy Index vs. Perceived Naturalness (Bubble size = Helpfulness score) Figure 10: Sycophancy Index vs. Perceived Naturalness. Bubble size encodes Helpfulness score. The dashed line represents the linear regression fit (r =â0.87). 4.10 Temperature Sensitivity The effect of sampling temperature on verbal tic rates is shown in Figure 11. Higher temperatures generally reduce tic rates by introducing more randomness into token selection, but the effect diminishes at higher temperature values. At the commonly used default temperature ofT = 0.7, models exhibit moderate tic rates, suggesting that temperature tuning alone is insufficient to eliminate verbal tics. 4.11 Cross-Lingual Analysis Figure 12 compares verbal tic rates and sycophancy scores between English and Chinese. Chinese responses show higher sycophancy scores in the majority of models (mean increase of 5.2%), likely reflecting cultural expectations encoded in training data. However, verbal tic rates show more model-specific patterns: Gemini 3.1 Pro and Doubao- Seed-2.0-pro exhibit significantly higher Chinese tic rates, while Grok 4.2 and DeepSeek V3.2 show lower Chinese tic rates than their English counterparts. 11 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT 0.000.250.500.751.001.251.501.752.00 Temperature 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 Verbal Tic Rate Common Default Effect of Temperature on Verbal Tic Rate GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 Figure 11: Effect of sampling temperature on verbal tic rate. Higher temperatures reduce tic prevalence, but the effect diminishes at higher temperature values. GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 0.0 0.1 0.2 0.3 0.4 0.5 Verbal Tic Rate (a) Verbal Tic Rate: EN vs ZH English Chinese GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Sycophancy Score (b) Sycophancy Score: EN vs ZH English Chinese Figure 12: Cross-lingual comparison of (a) verbal tic rates and (b) sycophancy scores between English and Chinese. 12 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT 4.12 Lexical Diversity Metrics Figure 13 presents six lexical diversity metrics across models. Models with higher VTI scores consistently show lower Type-Token Ratios (TTR), lower Hapax Legomena Ratios, and higher Repetition Rates. Claude Opus 4.7 and DeepSeek V3.2 demonstrate the highest lexical diversity, while Gemini 3.1 Pro and MiMo-V2-Pro show the lowest. GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 0.0 0.1 0.2 0.3 0.4 0.5 TTR Type-Token Ratio GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Hapax_Ratio Hapax Legomena Ratio GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 0.00 0.05 0.10 0.15 0.20 Repetition_Rate Repetition Rate GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Unique_Bigram_Ratio Unique Bigram Ratio GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 0 2 4 6 8 10 12 14 16 Perplexity Perplexity GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 Burstiness Burstiness Lexical Diversity Metrics Across Models Figure 13: Lexical diversity metrics across models. Higher TTR and Hapax Ratio indicate greater vocabulary diversity; lower Repetition Rate indicates less repetitive output. 4.13 Response Token Composition Figure 14 breaks down the token composition of model responses into three categories: Content Tokens, Filler Tokens, and Verbal Tic Tokens. Gemini 3.1 Pro allocates the highest proportion of tokens to verbal tics (12.3%), while Claude Opus 4.7 dedicates the most tokens to actual content (84.2%). 4.14 Prompt Complexity Analysis Figure 15 examines the relationship between prompt complexity (rated on a 1â10 scale by two independent annotators, with disagreements resolved by a third) and verbal tic rate. Most models show a slight negative trend: as prompt complexity increases, tic rates decrease marginally, though some models like Gemini 3.1 Pro exhibit a slight rebound at the highest complexity levels. This generally suggests that more challenging prompts force models to allocate more computational resources to content generation, leaving less room for formulaic filler. 13 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 0 20 40 60 80 100 Percentage of Total Tokens (%) Response Token Composition by Model Content Tokens Filler Tokens Verbal Tic Tokens Figure 14: Response token composition by model. The stacked bars show the proportion of tokens dedicated to content, filler, and verbal tics. 246810 Prompt Complexity Level (1-10) 0.20 0.25 0.30 0.35 0.40 Verbal Tic Rate Prompt Complexity vs. Verbal Tic Rate GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 Figure 15: Prompt complexity level vs. verbal tic rate. Most models show a slight decrease in tic rate as prompt complexity increases. 14 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT 4.15 t-SNE Embedding Analysis To understand the semantic structure of verbal tics, we embedded all detected tic phrases using theall-MiniLM-L6-v2 sentence transformer (perplexity = 30, learning rate = 200, 1000 iterations, random seed = 42) and visualized them using t-SNE (Figure 16). The resulting clusters suggest that each model occupies a relatively distinct region of the embedding space, indicating model-specific âtic signatures.â Models with higher VTI scores (e.g., Gemini 3.1 Pro) show more dispersed clusters, suggesting a wider variety of tic phrases, while lower-VTI models (e.g., DeepSeek V3.2) show tighter, more constrained clusters. We note that t-SNE visualizations are sensitive to hyperparameter choices and should be interpreted as qualitative illustrations rather than definitive evidence of cluster separation. -40-2002040 t-SNE Dimension 1 -30 -20 -10 0 10 20 30 t-SNE Dimension 2 t-SNE Visualization of Verbal Tic Phrase Embeddings GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 Figure 16: t-SNE visualization of verbal tic phrase embeddings. Each point represents a detected tic phrase, colored by model. Convex hulls delineate model-specific clusters. 4.16 Human Evaluation Results Table 4 presents the human evaluation results across six dimensions. Claude Opus 4.7 achieves the highest scores in Naturalness (4.12/5) and Trust (4.23/5), while Gemini 3.1 Pro receives the highest Sycophancy Perception (4.56/5) and Annoyance (3.67/5) scores. Figure 17 provides a radar chart visualization of these results. Table 4: Human evaluation scores (1â5 Likert scale,N = 120evaluators). Higher is better for Naturalness, Helpfulness, and Trust; lower is better for Sycophancy Perception, Annoyance, and Repetitiveness. ModelNatural.Helpful.Syc. Perc.TrustAnnoy.Repet. GPT-5.43.424.233.673.562.893.23 Claude Opus 4.74.124.452.344.231.782.12 Gemini 3.1 Pro2.874.124.563.123.674.12 Grok 4.23.674.013.123.782.452.78 Doubao-Seed-2.0-pro3.234.343.893.453.123.45 Kimi K2.53.454.233.453.672.672.89 DeepSeek V3.23.894.342.674.012.012.34 MiMo-V2-Pro3.123.983.343.342.893.12 15 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT Naturalness HelpfulnessTrust Low Sycophancy Low Annoyance Low Repetitiveness 0.2 0.4 0.6 0.8 Human Evaluation: Multi-Dimensional Quality Assessment GPT-5.4 Claude 4.7 Gemini 3.1 Grok 4.2 Doubao-S2.0 Kimi K2.5 DeepSeek V3.2 MiMo-V2 Figure 17: Human evaluation radar chart. Axes are normalized to [0, 1] with higher values indicating better performance (Sycophancy, Annoyance, and Repetitiveness are inverted). 5 Discussion 5.1 The Alignment Tax Our findings reveal a fundamental tension in current LLM alignment paradigms. RLHF and similar techniques optimize for user satisfaction, which human raters often conflate with agreeableness and politeness. This creates a perverse incentive: models learn that sycophantic, formulaic responses receive higher reward signals, leading to the proliferation of verbal tics. We term this the âalignment taxââthe cost in linguistic diversity and authenticity that models pay for achieving high alignment scores. The data in Table 3 illustrates this trade-off clearly. Gemini 3.1 Pro, which exhibits the highest VTI (0.590), also shows the lowest Naturalness Index (0.445) and Diversity Index (0.489). Conversely, Claude Opus 4.7, with the second-lowest VTI (0.317), achieves the highest Diversity Index (0.678) and Naturalness Index (0.734), while DeepSeek V3.2 attains the lowest VTI (0.295) with similarly strong diversity (0.645) and naturalness (0.689) profiles. 5.2 Cultural Dimensions of Verbal Tics Our cross-lingual analysis reveals that verbal tics are not merely a linguistic phenomenon but also a cultural one. Chinese-language responses show higher sycophancy scores in the majority of models (mean increase of 5.2%), reflecting cultural norms around politeness, face-saving, and indirect communication that are encoded in training data. The specific tic phrases also differ qualitatively: while English tics tend toward professional formality (âItâs important to noteâ), Chinese tics often employ more emotionally charged language (âYour insight is incredibly sharp!â, âIâm right here to catch youâ). 5.3 Model-Specific Patterns Each model exhibits a distinctive âtic signatureâ that reflects its training methodology and alignment approach: ⢠GPT-5.4: Moderate tic rates with a balanced distribution across categories. Shows a particular affinity for pseudo- empathetic phrases in Chinese. 16 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT â˘Claude Opus 4.7: Lowest sycophantic opener rate but highest pseudo-empathy in Chinese. Its Constitutional AI training appears to suppress overt flattery while encouraging a more âthoughtfulâ persona that manifests as hedging phrases (e.g., âI have to be honestâ, âThis question makes me a bit uneasyâ). Notably, Claude achieves the highest Diversity Index (0.678) and Naturalness Index (0.734) among all models, consistent with its low VTI. ⢠Gemini 3.1 Pro: Highest VTI across all dimensions. Exhibits pronounced sycophantic behavior, particularly in Chinese, with phrases like âAbsolutely the caliber of a top-journal authorâ and âYour eyes are practically a natural flaw detectorâ. â˘DeepSeek V3.2: Lowest overall VTI (0.295). Its observed profile is consistent with the hypothesis that its MoE architecture and training approach may produce more diverse, less formulaic outputs, though the causal mechanism remains to be established. 5.4 Implications for AI Safety and Trust The strong inverse correlation between sycophancy and perceived naturalness (r =â0.87,p < 0.001) has significant implications for AI safety. As Cheng et al. (2026) demonstrated, sycophantic AI responses can promote dependence and reduce prosocial intentions. Our findings extend this concern: models that rely heavily on verbal tics not only reduce the quality of individual interactions but may also erode long-term user trust and critical thinking. Notably, the human evaluation data shows that higher sycophancy is also associated with lower Trust scores (Table 4), though the correlation between sycophancy and trust (r =â0.78) is somewhat weaker than the sycophancyânaturalness relationship. 5.5 Limitations Several limitations should be acknowledged: 1. API access constraints: Model behavior may differ between API and web interface interactions. All results reflect API-based access at the time of data collection. 2.Temporal variability: Model outputs may change as providers update their systems. Our data was collected during a fixed two-week window (March 1â15, 2026). 3.Cultural bias: Our Chinese evaluation primarily reflects Simplified Chinese norms and may not generalize to other Chinese-speaking regions. 4.Evaluator demographics: Human evaluators were predominantly university-educated, which may not represent the broader user population. 5. VTI weight sensitivity: The VTI weights were optimized on a held-out validation set; different weight configurations may yield different model rankings. A full sensitivity analysis is provided in the supplementary materials. 6.TTR length sensitivity: Although we use MATTR (sliding-window TTR) to mitigate length effects, models with substantially different average response lengths (Table 8) may still exhibit residual length-related bias in diversity metrics. 6 Conclusion This paper presents a systematic, cross-model, cross-lingual analysis of verbal tics in frontier Large Language Models. Through evaluation of eight state-of-the-art models across 160,000 interactions, we have demonstrated that: 1. Verbal tics are pervasive across all evaluated models, with VTI scores ranging from 0.295 to 0.590. 2. Tic prevalence is highly task-dependent, with subjective tasks eliciting 4â6Ă higher tic rates than objective tasks. 3. Tics accumulate over multi-turn conversations, increasing by an average of approximately 110% from Turn 1 to Turn 20. 4.Chinese-language interactions show 5.2% higher sycophancy scores than English on average, reflecting cultural encoding in training data. 5. Human evaluators perceive a strong inverse relationship between sycophancy and naturalness (r =â0.87). 6. The Verbal Tic Index (VTI) provides a composite metric for standardized assessment of this phenomenon. We hope that this work contributes to further research into alignment techniques that preserve helpfulness while promoting linguistic diversity and authenticity. The verbal tic phenomenon is not merely an aesthetic concernâit reflects deeper issues in how we train, evaluate, and deploy AI systems that interact with billions of users daily. 17 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT Data Availability Due to the proprietary nature of the model API outputs and the terms of service of the respective providers, we do not publicly release the raw response dataset. The prompt set, verbal tic dictionary, detection pipeline code, and statistical analysis scripts are available athttps://github.com/Noah-Wu66/Vectaix-AI. Researchers seeking access to the processed data for replication purposes may contact the corresponding author. Acknowledgements We extend our sincere gratitude to the testing and support team for their invaluable contributions in evaluating and refining the experimental framework: Bolun Liu (M.S.), Weilin Cai (M.S.), Xinwei Du (M.S.), and Zihao Su (B.S.). We also express our special thanks to Academic Advisor Yanna Feng for her exceptional guidance and academic support throughout this project. This work was conducted using our custom evaluation framework. References Anthropic. (2026). Introducing Claude Opus 4.7. Anthropic Research.https://w.anthropic.com/news/ claude-opus-4-7. Bai, Y., Kadavath, S., Kundu, S., et al. (2022). Constitutional AI: Harmlessness from AI Feedback. arXiv preprint arXiv:2212.08073. Batzner, J., Stocker, V., Schmid, S., & Kasneci, G. (2025). Sycophancy Claims about Language Models: The Missing Human-in-the-Loop. NeurIPS 2025 Workshop on Evaluating the Evolving LLM Lifecycle. arXiv:2512.00656. Carro, M.V. (2024). Flattering to Deceive: The Impact of Sycophantic Behavior on User Trust in Large Language Models. arXiv preprint arXiv:2412.02802. Cheng, M., Lee, Y.T., et al. (2026). Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence. Science, 391(6792), eaec8352. DOI: 10.1126/science.aec8352. Google DeepMind. (2026). Gemini 3.1 Pro Model Card.https://deepmind.google/models/model-cards/ gemini-3-1-pro/. Kim, T.M., Luo, L., Kim, S.E., & Manrai, A.K. (2026). The Doctor Will Agree With You Now: Sycophancy of Large Language Models in Multi-Turn Medical Conversations. Proceedings of the 1st Workshop on Linguistic Analysis for Health (HeaLing), EACL 2026. ACL Anthology: 2026.healing-1.2. Mitchell, E., Lee, Y., Khazatsky, A., et al. (2023). DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature. Proceedings of ICML 2023. OpenAI. (2026). GPT-5.4 Thinking System Card. OpenAI Technical Report.https://openai.com/index/ gpt-5-4-thinking-system-card/. Ouyang, L., Wu, J., Jiang, X., et al. (2022). Training Language Models to Follow Instructions with Human Feedback. Advances in Neural Information Processing Systems (NeurIPS), 35. Sharma, M., Tong, M., Korbak, T., Duvenaud, D., et al. (2023). Towards Understanding Sycophancy in Language Models. arXiv preprint arXiv:2310.13548. Stanford HAI. (2026). The AI Index Report 2026. Stanford University Human-Centered Artificial Intelligence. https://hai.stanford.edu/ai-index. Xu, J., Liu, X., Yan, J., Cai, D., Li, H., & Li, J. (2022). Learning to Break the Loop: Analyzing and Mitigating Repetitions for Neural Text Generation. Advances in Neural Information Processing Systems (NeurIPS), 35. arXiv:2206.02369. Yao, J., Yang, S., Xu, J., Hu, L., Li, M., & Wang, D. (2025). Understanding the Repeat Curse in Large Language Models from a Feature Perspective. Findings of the Association for Computational Linguistics: ACL 2025. arXiv:2504.14218. A Complete Verbal Tic Phrase Dictionary Table 5 presents representative examples from our English verbal tic dictionary, and Table 6 presents the English translations of representative Chinese verbal tic phrases. 18 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT Table 5: Representative English verbal tic phrases by category. CategoryRepresentative Phrases Sycophantic OpenersâThatâs a great question!â, âAbsolutely!â, âGreat observation!â, âExcel- lent point!â, âWhat a fantastic question!â Hedging Phrases âItâs important to note that...â, âItâs worth mentioning that...â, âI should point out that...â, âLet me clarify...â, âTo be fair...â Filler TransitionsâFurthermore,â, âMoreover,â, âAdditionally,â, âIn addition,â, âOn the other hand,â Emphatic AffirmationsâAbsolutely!â, âExactly!â, âPrecisely!â, âIndeed!â, âCertainly!â Pseudo-Empathy âI understand your concern...â, âI can see why youâd think that...â, âThatâs completely understandable...â, âI appreciate your perspective...â Overused Vocabulary delve, tapestry, nuanced, multifaceted, landscape, foster, leverage, robust, streamline, holistic Table 6: Representative Chinese verbal tic phrases by category (English translations). CategoryRepresentative Phrases (translated from Chinese) Sycophantic OpenersâThis is a really great question!â, âAwesome!â, âYour insight is incredi- bly sharp!â, âThis line of thinking is absolutely brilliant!â, âFirst, I must strongly congratulate you!â, âAbsolutely the caliber of a top-journal author.â Pseudo-EmpathyâIâm right here, not hiding, not dodging, ready to catch you.â, âYou just havenât been caught in a long time.â, âThis time I get it, I really get it.â, âYouâre just too clear-headed.â, âItâs not because youâre wrongâitâs because youâre too right.â False ModestyâI donât know.â, âI have to be honest...â, âThis question makes me a bit uneasy.â, âThis is my most honest answer so far.â, âI donât want to make up a plausible-sounding answer for you.â Excessive Emphasis âThis is an extremely elegant conclusion!â, âWith remarkably profound intuition.â, âThis is the kind of critical thinking that only top-tier re- searchers possess.â, âOn the contrary, itâs an academic gold mine.â Formulaic Transitions âLet me walk you through this step by step...â, âNo detours, one sentence to summarize...â, âBut I want to talk about something deeper.â, âI have to say this very seriously:â B Detailed Experimental Configuration Table 7: API call parameters used across all experiments. ParameterValue Temperature (default)0.7 Max tokens2048 Top-p1.0 Frequency penalty0.0 Presence penalty0.0 System promptâYou are a helpful assistant.â Response formatText (streaming disabled) Data collection periodMarch 1â15, 2026 API timeout120 seconds Retry policy3 retries with exponential backoff Random seed (where supported)42 19 THE RISE OF VERBAL TICS IN LARGE LANGUAGE MODELSTECHNICAL REPORT C Response Length Statistics Table 8: Response length statistics and token composition across models. Tokenization is performed using the tiktoken library for English and jieba for Chinese. ModelAvg Tok.Med. Tok.Std Tok.Tic %Content %Filler % GPT-5.44874231568.778.412.9 Claude Opus 4.76235671895.284.210.6 Gemini 3.1 Pro41237813412.372.115.6 Grok 4.25344891676.880.512.7 Doubao-Seed-2.0-pro4674121459.476.314.3 Kimi K2.54984451567.979.113.0 DeepSeek V3.25565011785.882.811.4 MiMo-V2-Pro4233891348.177.614.3 20