Paper deep dive
System One Models for Wireless Decision-Making:Applications and Performance Evaluation
Masoud Rahimi, S. M. Matin Alemohammad, Hamid Behroozi, Mahdi Nouri
Intelligence
Status: succeeded | Model: Gemma-4-26B-A4B | Prompt: intel-v1 | Confidence: 90%
Last extracted: 10/6/2026, 11:19:42 AM
Summary
This paper investigates System-One models as lightweight alternatives to Large Language Models (LLMs) for wireless decision-making tasks requiring bounded, low-latency actions. The authors formalize System-One models as schema-conditioned probabilistic decision interfaces that output distributions over explicit decision spaces rather than generating tokens autoregressively. Using the 'Jev' implementation, they benchmark performance against LLMs and conventional baselines in scenarios like receive-antenna selection and intent-conditioned RAN slicing. Results show significant latency reductions (up to 8.5x) with acceptable trade-offs in decision quality, positioning System-One models as a promising interface for latency-sensitive, intent-driven wireless control.
Entities (8)
Relation Signals (6)
Jev → implements → System-One Models
confidence 95% · Using Jev as a System-One implementation, we benchmark decision quality...
Jev → achieveslatencyreductionin → Receive-Antenna Selection
confidence 92% · In receive-antenna selection, Jev delivers up to an 8.5x reduction in median response latency
Jev → achieveslatencyreductionin → RAN Slicing
confidence 92% · in intent-conditioned RAN slicing, Jev achieves utility comparable to the evaluated LLMs while providing more than a 3.5x reduction in median response latency.
System-One Models → isalternativeto → Large Language Models
confidence 90% · This paper investigates System-One models... as a lightweight alternative for wireless decision-making... relative to the evaluated LLMs
System-One Models → appliesto → Physical-layer control
confidence 88% · identify their applicability across physical-layer control, radio resource management, mobility, network slicing, and network operations
System-One Models → usestrainingmethod → RLCD
confidence 85% · Jev... use... a calibration-oriented training method termed Reinforcement Learning for Calibrated Decisions (RLCD)
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:Many wireless control tasks require repeated selection of a single action from a finite feasible set under stringent latency and reliability constraints. While large language models (LLMs) have recently emerged as general-purpose decision engines, their autoregressive generation mechanism is not naturally aligned with such bounded control problems. This paper investigates System-One models, which directly learn probability distributions over explicitly defined decision spaces, as a lightweight alternative for wireless decision-making. We formalize their decision structure and learning objective, identify their applicability across physical-layer control, radio resource management, mobility, network slicing, and network operations, and evaluate their practical behavior through representative wireless case studies. Using Jev as a System-One implementation, we benchmark decision quality and client-observed latency against generative LLMs and conventional baselines. In receive-antenna selection, Jev delivers up to an 8.5x reduction in median response latency relative to the evaluated LLMs, although this gain comes with a loss in decision quality compared with stronger task-specific alternatives. More notably, in intent-conditioned RAN slicing, Jev achieves utility comparable to the evaluated LLMs while providing more than a 3.5x reduction in median response latency. Complementary evidence from edge-service orchestration further shows that faster decisions do not necessarily translate into lower end-to-end service latency. These results expose a fundamental quality/latency tradeoff and position System-One models not as replacements for numerical optimization, but as a promising decision interface for latency-sensitive, bounded, and intent-driven wireless control.
Tags
Links
- Source: https://arxiv.org/abs/2610.04345v1
- Canonical: https://arxiv.org/abs/2610.04345v1
Trouble viewing inline? Open PDF directly →
Full Text
52,893 characters extracted from source content.
Expand or collapse full text
System One Models for Wireless Decision-Making: Applications and Performance Evaluation Masoud Rahimi S. M. Matin Alemohammad Hamid Behroozi Mahdi Nouri †thanks: The authors are with the Department of Electrical Engineering, Sharif University of Technology, Tehran 1458889694, Iran (e-mail: masoud.rahimi@sharif.edu; matin.alemohammad79@sharif.edu). Abstract Many wireless control tasks require repeated selection of a single action from a finite feasible set under stringent latency and reliability constraints. While large language models (LLMs) have recently emerged as general-purpose decision engines, their autoregressive generation mechanism is not naturally aligned with such bounded control problems. This paper investigates System-One models, which directly learn probability distributions over explicitly defined decision spaces, as a lightweight alternative for wireless decision-making. We formalize their decision structure and learning objective, identify their applicability across physical-layer control, radio resource management, mobility, network slicing, and network operations, and evaluate their practical behavior through representative wireless case studies. Using Jev as a System-One implementation, we benchmark decision quality and client-observed latency against generative LLMs and conventional baselines. In receive-antenna selection, Jev delivers up to an 8.5x reduction in median response latency relative to the evaluated LLMs, although this gain comes with a loss in decision quality compared with stronger task-specific alternatives. More notably, in intent-conditioned RAN slicing, Jev achieves utility comparable to the evaluated LLMs while providing more than a 3.5x reduction in median response latency. Complementary evidence from edge-service orchestration further shows that faster decisions do not necessarily translate into lower end-to-end service latency. These results expose a fundamental quality/latency tradeoff and position System-One models not as replacements for numerical optimization, but as a promising decision interface for latency-sensitive, bounded, and intent-driven wireless control. Index Terms: System 1 models, 6G, wireless intelligence, decision models, large language models, agentic AI, uncertainty calibration, radio resource management. I Introduction Artificial intelligence is becoming an integral component of future wireless networks, spanning physical-layer processing, radio-resource management, mobility, network operations, and service orchestration [1, 2]. Many of these functions differ fundamentally from open-ended generative tasks: they repeatedly map an observed network state to a bounded operational decision, such as selecting a modulation-and-coding scheme (MCS), beam, serving cell, scheduling policy, antenna subset, or remediation action. Such decisions must often be produced under tight latency, reliability, and feasibility requirements. Wireless intelligence has traditionally addressed these problems through model-based optimization and task-specific learning. Optimization provides explicit objectives and constraints but can require iterative numerical procedures whose online cost grows with problem complexity. Learned policies can amortize this computation and provide fast inference, but they are commonly designed for a fixed task, state representation, and action space [3, 4, 5, 6]. Foundation models offer a different objective: reuse learned representations and knowledge across multiple tasks rather than training an independent model for every function [7, 8]. Recent work has explored this direction through telecom-specific language models [9], LLM adaptation for networking tasks [10], and wireless-native foundation models operating on RF and channel data [11, 12, 13, 14]. However, the appropriate role of an LLM depends on the structure of the task. Natural-language intent interpretation, knowledge retrieval, planning, and troubleshooting can benefit directly from generative language models. In contrast, many radio and network-control tasks ultimately require a decision from an explicitly admissible set. Autoregressive generation is then an indirect interface: an answer is constructed token by token and subsequently parsed, validated, and mapped to an executable action. This distinction has already appeared in networking research. NetLLM, for example, uses task-specific networking heads that bypass free-form generation and directly produce valid task outputs [10]. These observations motivate a broader question: when the desired output is a bounded decision rather than generated content, should decision inference itself be the native model interface. Recently, the term System-One model has been introduced for decision-oriented models that replace free-form sequence generation with typed probabilistic queries over a declared answer space [15, 16]. At the interface level, such a model receives an observed state s, a decision query q, and an admissible answer set A, and returns a probability distribution pθ(Y∣,q,),Y∈,p_θ(Y ,q,A), Y , (1) rather than a free-form token sequence. The distinction is not merely that the output is categorical. The query and admissible outcomes are themselves conditioning variables, enabling a shared model, in principle, to address different decision schemas without a separately trained output head for every task. The interface also guarantees schema-valid outputs and exposes probabilities that can support confidence-aware execution, abstention, or escalation. The first public implementation, Jev, is reported by TypeSafe to use a new model architecture, a parallel sampler, and a calibration-oriented training method termed Reinforcement Learning for Calibrated Decisions (RLCD) [15]. By eliminating token-by-token output generation, it is explicitly designed for low-latency machine decision making; TypeSafe reports 70–500 ms end-to-end response times and 40–200× speedups over frontier LLMs on its System-One-oriented workloads. These figures are vendor-reported and workload dependent, but they motivate evaluating the computational advantage directly in wireless settings. AnyJev provides a complementary open-source route by converting pretrained LLMs into System-One-style typed decision interfaces without fine-tuning their backbone weights. It obtains bounded decisions through non-generative readouts and supports zero-label debiasing as well as lightweight, per-question calibration or closed-form decision heads [17]. Thus, the System-One interface can be separated from the architecture used to implement it. Preliminary third-party evidence has also been reported from a customer-side evaluation by Deel. On two production decision tasks, the published results show approved-answer matching increasing from 70% with a frontier LLM to 97% with Jev, and expense-categorization agreement with human reviewers increasing from 50% to 86%, as summarized in Fig. 1 [18]. Because the evaluated datasets, sample sizes, comparator models, and prompting configurations have not been publicly documented, these figures should be interpreted as customer-reported evidence rather than an independently reproducible benchmark. Fig. 1: Reported quality on two production decision tasks evaluated by Deel. The first panel reports approved-answer matching accuracy; the second reports expense-categorization agreement with human reviewers. Results were published by TypeSafe AI [18]. Wireless networks provide a natural testbed for this model class because many control functions combine repeated state-conditioned inference with bounded actions. Candidate examples include link adaptation, beam and antenna selection, handover, slice admission, scheduling-policy selection, energy-control modes, and fault-response routing. The fit is not universal. High-dimensional continuous outputs such as channel estimates, arbitrary precoding matrices, or waveforms are better handled by specialized estimators, regression or generative models, or optimization unless they are decomposed into bounded sub-decisions. Likewise, hard physical and protocol constraints should remain under deterministic or optimization-based control rather than being delegated solely to a learned probabilistic model. This perspective also motivates a hybrid System-One/System-Two architecture for wireless intelligence. Routine decisions with sufficiently high confidence can follow a fast decision path, whereas ambiguous, out-of-distribution, or high-risk states can trigger additional measurements or be escalated to an optimizer, simulator, or LLM-based reasoning agent. The resulting question is therefore not whether System-One models should replace existing wireless algorithms, but where a shared decision-native model provides measurable value in decision quality, latency, calibration, robustness, and cross-task reuse, and how it should interact with established model-based and learning-based methods. The main contributions of this article are as follows: • We formalize System-One models as schema-conditioned probabilistic decision models and clarify their relationship to conventional classifiers, task-specific learned policies, and autoregressive LLMs. • We identify the structural properties that make a wireless task suitable for this interface and organize representative opportunities across PHY, MAC/RRM, mobility, and network management. • We evaluate the decision-model paradigm through receive-antenna selection and intent-conditioned RAN slicing, and complement these experiments with independently reported intent-driven edge-service orchestration results. The comparison considers conventional baselines and generative LLMs and examines decision quality, latency, and end-to-end utility. The remainder of the article is organized as follows. Section I develops the System-One decision interface and learning formulation. Section I examines candidate wireless applications and suitability criteria. Section IV presents the evaluation methodology and case studies, while section V discusses open problems in wireless-native decision models, calibration, safety, and deployment. I System-One Models: Functional Structure and Learning Objectives As introduced in equation 1, given an observed state s, a decision query q, and a finite admissible answer space A, the model provides a conditional distribution pθ(Y∣s,q,)p_θ(Y s,q,A) over the admissible outcomes. Here, s contains the information available for decision making, q specifies the decision criterion, and A defines the allowable outputs. The primary model output is therefore a probability distribution over decisions rather than a free-form sequence of generated tokens. This formulation defines the model at the level of its probabilistic interface and task structure. It does not depend on a particular neural backbone, prediction head, or optimization algorithm. Different implementations may realize the same conditional decision distribution using different architectures and training procedures. The answer space can encode categorical, binary, or ordered decisions. For a categorical decision with =a1,…,aKA=\a_1,…,a_K\, the predicted answer is a^=argmaxak∈pθ(Y=ak∣s,q,). a= a_k \;p_θ(Y=a_k s,q,A). (2) Binary decisions are simply the special case K=2K=2. The same formulation also covers ordered decision spaces. If each outcome aka_k is associated with an ordered numerical level vkv_k, let V denote the corresponding discrete-valued random variable, i.e., V=vkV=v_k when Y=akY=a_k. The predictive distribution then induces the scalar summary r^=[V∣s,q,]. r=E[V s,q,A]. (3) This quantity is the conditional expectation over the explicitly declared numerical levels. It therefore remains tied to the specified decision schema and should not be interpreted as an unrestricted continuous regression output. Learning is formulated as conditional distribution learning over decision spaces. Let x=(,q,),x=(s,q,A), (4) and let Y denote the target decision outcome. The learning problem is θ⋆=argminθ(x,Y)∼[ℓ(pθ(⋅∣x),Y)],θ = _θE_(x,Y) [ \! (p_θ(· x),Y ) ], (5) where ℓ evaluates the quality of the reported predictive distribution. Strictly proper scoring rules provide a natural objective for this setting because their expected value is optimized when the reported distribution matches the underlying conditional distribution. Cross-entropy and the Brier score are standard examples. For instance, cross-entropy with a target distribution t is ℒCE=−(x,)∼[∑k=1Ktklogpθ(Y=k∣x)],L_CE=-E_(x,t) [ _k=1^Kt_k p_θ(Y=k x) ], (6) with hard-label training obtained as the special case tk=[k=Y]t_k=1[k=Y]. The probabilistic learning objective is distinct from the optimization procedure used to fit the model. Supervised likelihood-based training, calibration-oriented objectives, reinforcement learning, and hybrid procedures provide different mechanisms for learning the conditional distribution pθ(Y∣,q,)p_θ(Y ,q,A) [15, 16]. These procedures therefore represent alternative learning strategies rather than defining characteristics of the System-One model class. The distinction from a conventional fixed-label classifier lies primarily in how the prediction problem is specified. A conventional classifier typically represents pθ(Y∣),p_θ(Y ), (7) where the task and label space are largely fixed during model development. In the System-One abstraction, the query q and admissible answer space A are explicit parts of the model input. The same pretrained model can therefore perform different tasks without task-specific fine-tuning: changing q and A changes the decision problem and its output space while keeping θ fixed. Unlike a conventional fixed-label classifier, it does not require a newly trained output head for each task. This cross-task reuse is a central capability, although accuracy and calibration on each new task must still be evaluated. The distinction from an autoregressive language model is equally fundamental at the level of output factorization. An autoregressive language model represents a distribution over token sequences, pψ(w1:L∣X)=∏ℓ=1Lpψ(wℓ∣X,w<ℓ),p_ψ(w_1:L X)= _ =1^Lp_ψ(w_ X,w_< ), (8) and constructs its output through sequential token prediction. A System-One interface, in contrast, exposes a probability distribution directly over the declared decision space and uses that distribution to select or score admissible actions. Autoregressive language models can also perform classification and structured prediction, but in that case the decision is expressed through, or derived from, a token-generation process rather than being the native modeled output space. Finally, probabilistic output must be distinguished from calibration. Normalization guarantees only that the reported values form a valid probability distribution; it does not guarantee that those probabilities correspond to empirical frequencies of correctness. A model can therefore produce normalized probabilities while remaining systematically over-confident or under-confident. Calibration is an empirical property of the learned predictor relative to a data distribution and deployment regime, and it must be assessed independently of the structural definition of a System-One model. I System-One Decision Opportunities in Wireless Networks Having established the probabilistic formulation of System-One models, we now examine where this decision interface may be useful in wireless networks. The opportunities span PHY-level control, MAC and radio resource management, mobility, and network operations. We first identify the characteristics of a suitable decision task and then use these criteria to organize the candidate applications and select representative case studies. I-A Characteristics of Suitable Wireless Decisions Not every wireless function is an appropriate target for a System-One model. Suitability is determined less by the protocol layer than by the structure, timing, and risk profile of the decision. The following characteristics provide a practical screening framework. 1) A bounded and explicit decision space: The task should admit a finite set of meaningful alternatives, such as an MCS index, beam identifier, serving cell, scheduling mode, alarm class, or sleep/wake command. This structure allows the model to return a normalized distribution over feasible outcomes and enables the controller to reject out-of-schema actions by construction. Tasks whose native output is a high-dimensional continuous object, such as an unconstrained precoding matrix or waveform, are less direct candidates unless the decision is discretized, factorized into several bounded queries, or combined with a downstream optimizer. 2) Repeated state-conditioned decisions: The same decision semantics should recur across many network states. Wireless control commonly has this form: channel, traffic, queue, mobility, and load measurements are observed repeatedly, and an action is selected from a well-defined set [6]. This favors amortized inference, in which computational effort is shifted to training and decisions are produced quickly online. In contrast, one-off engineering tasks that require open-ended design, extensive explanation, or long-horizon decomposition are better suited to optimization tools, simulators, or deliberative agents. A System-One model may still support such workflows, but only as a bounded decision component. 3) A compatible end-to-end control timescale: The relevant requirement is not model latency alone, but the complete delay from state acquisition through preprocessing, inference, validation, signaling, and actuation. O-RAN identifies near-real-time control loops operating roughly between 1010 ms and 11 s and non-real-time loops above 11 s, while sub-1010-ms loops remain closely tied to embedded RAN processing and hardware constraints [19]. System-One models are therefore initially more plausible for decisions with millisecond-to-second budgets than for symbol-level processing. Tighter loops remain possible only if the complete implementation, including the input interface and safety checks, satisfies the deadline. 4) Uncertainty that changes the downstream action: Probabilistic output is most valuable when confidence determines whether a decision is executed, verified, deferred, or passed to a fallback method. Uncertainty-aware interference management, for example, has been used to switch between a fast learned solution and a conventional algorithm when the learned prediction is insufficiently trustworthy [20]. A suitable System-One task should similarly define what the controller does with low confidence; otherwise, returning a probability distribution provides little operational benefit. This requirement also makes calibration under deployment shift as important as average accuracy. 5) Reusable structure across related decision schemas: The strongest motivation for a shared decision model arises when states and decision semantics vary across cells, configurations, or objectives, but retain common structure. Conventional DRL policies may require substantial retraining when state or action spaces change, reducing sample efficiency and limiting generalization [21]. A System-One model is particularly relevant when the answer set or decision query must be specified at runtime. However, schema flexibility should not be assumed to imply zero-shot competence: transfer to new queries, environments, and action sets must be measured explicitly. 6) Enforceable constraints and a safe fallback: Many wireless decisions are subject to hard power, spectrum, reliability, fairness, and protocol constraints. A probabilistic model should not be solely responsible for enforcing requirements that can be checked deterministically. The selected action should pass through a feasibility layer, and the system should retain an optimization-based, rule-based, or human fallback for out-of-distribution or high-risk states. This hybrid view is consistent with recent knowledge-driven wireless AI frameworks, which combine learned components with physical models and optimization structure [22]. No single characteristic is sufficient by itself. Suitability increases when a task combines a bounded action space, repeated state-conditioned execution, an achievable end-to-end deadline, and a meaningful use for calibrated confidence, while also permitting explicit constraint checking and a credible fallback. These criteria are applied across the protocol stack in the following subsections and used to select the representative case studies in section IV. I-B PHY-Level Decisions At the PHY layer, the most direct opportunities are not arbitrary signal reconstruction problems, but discrete configuration decisions made from channel measurements, feedback, and recent transmission outcomes. Link adaptation is a representative example. In NR, the transmitter selects an MCS from standardized tables that jointly determine the modulation order and target code rate [23]. Let tLAs^LA_t collect features such as effective SINR, CQI, HARQ acknowledgements, rank and precoding indicators, Doppler or mobility information, and recent BLER. With the admissible MCS set ℳM, a simplified reliability-constrained decision is mt⋆= m_t = argmaxm∈ℳrm[1−P^BLER(m∣tLA)] _m r_m\! [1- P_BLER(m ^LA_t) ] (9) s.t. .t. P^BLER(m∣tLA)≤ϵ, P_BLER(m ^LA_t)≤ε, where rmr_m is the spectral efficiency of MCS m and ϵε is the target BLER. Existing work has already shown that learning can complement outer-loop link adaptation when the mapping from compact feedback to BLER is difficult to model accurately [24]. A System-One model would return a distribution over ℳM, rather than only the maximizing index. This distribution could support conservative fallback to a lower MCS, top-k verification, or abstention when the state is far from the training distribution. The principal challenge is timing: the complete feature-construction, inference, and scheduling path must fit within the link adaptation deadline. Beam selection has a similar bounded structure when analog or hybrid beamforming uses a finite codebook ℬ=b1,…,bKBB=\b_1,…,b_K_B\. The state may contain partial CSI, received powers, position, angle estimates, blockage indicators, or previous beam measurements, and the action is a beam or beam pair. A natural target is bt⋆=argmaxb∈ℬR^(b∣tB),b_t = _b R(b ^B_t), (10) where R R denotes predicted rate or received power. Recent probability-map approaches explicitly predict the probability that each candidate is optimal and adapt the number of beams retained for subsequent training [25]. This is closely aligned with a typed decision distribution: the controller may select one beam when confidence is high or a small candidate subset whose cumulative probability exceeds a threshold when confidence is diffuse. Other bounded PHY decisions include antenna or spatial-layer selection, receiver-mode selection, numerology or bandwidth-part selection, and choosing among predefined detection, decoding, or channel-estimation procedures. However, raw channel estimation, waveform generation, and unconstrained precoder design produce continuous high-dimensional objects and are not direct System-One tasks. They require either a specialized estimator, a generative or regression model, or a decomposition in which the System-One model selects a mode, codebook entry, or solver configuration while another component computes the continuous parameters. I-C MAC, RRM, and Mobility Decisions MAC and radio resource management expose a large number of repeated decisions, but their suitability depends strongly on how the action space is represented. Let ts_t include per-user channel indicators, queue lengths, packet deadlines, slice requirements, interference, and cell load. Many resource-management objectives can be expressed abstractly as at⋆= a_t = argmax∑ua∈(t)Uu(t,a) _a (s_t) _uU_u\! (s_t,a ) (11) s.t. .t. gi(t,a)≤0,i=1,…,I, g_i(s_t,a)≤ 0, i=1,…,I, where UuU_u captures rate, delay, or fairness utility and gig_i represents resource and QoS constraints. Directly treating every joint assignment of users, resource blocks, beams, and power levels as one categorical action is usually infeasible because the number of alternatives grows combinatorially. A more credible System-One formulation is hierarchical: select a scheduling rule, a user group, a slice-allocation profile, or a short list of candidate schedules, and then use a constrained scheduler or optimizer to produce the final allocation. RAN slicing illustrates this intermediate level. The state contains slice and UE metrics, while the action may select among scheduler policies or bounded resource-sharing profiles intended to satisfy service-level intents. Intent-aware learning has been evaluated for allocating scarce radio resources among slices with heterogeneous requirements [26]. A typed decision model could expose the probability of each scheduling profile and abstain when none is predicted to satisfy the declared intents. Continuous power control can be handled similarly by using a finite power codebook or by selecting a power-control policy; unrestricted per-resource-block power vectors should remain with a continuous controller. Mobility management provides an especially natural decision schema. For a UE with serving cell c0c_0 and candidate neighbors t=c1,…,cKC_t=\c_1,…,c_K\, the action set is tHO=stay at c0, hand over to c1,…,cK.A^HO_t=\stay at c_0, hand over to c_1,…,c_K\. (12) The state may combine serving and neighboring RSRP/RSRQ/SINR histories, UE velocity, cell load, QoS, and recent handovers. A useful decision objective is J(a)=[R¯(a)−λfF(a)−λpP(a)−λoO(a)],J(a)=E\! [ R(a)- _fF(a)- _pP(a)- _oO(a) ], (13) where F, P, and O denote handover failure, ping-pong, and outage costs. O-RAN traffic-steering work has demonstrated user-specific serving-cell selection using near-real-time data and learned control policies [27]. A System-One distribution over the current candidate cells would additionally make confidence explicit: high-confidence actions may be executed, whereas ambiguous cases can retain the serving cell, trigger additional measurements, or invoke a slower policy. I-D Network Management Decisions Network-management decisions operate on slower timescales and broader state descriptions than PHY or MAC control. The state may include KPI windows, alarm sequences, configuration changes, topology, traffic forecasts, security events, and service-level objectives. If a set of approved remediation actions is available, the problem can be represented as at⋆=argmina∈NM[ a_t = _a ^NME\! [ Csvc(a)+λeCenergy(a) C_svc(a)+ _eC_energy(a) (14) +λrCrisk(a)∣NMt], + _rC_risk(a) ^NM_t ], subject to coverage, capacity, security, and policy constraints. Possible actions include monitor, collect additional evidence, restart a component, rollback a configuration, throttle or isolate an entity, activate a predefined xApp/rApp, change an energy-saving mode, or escalate to an operator. Fault and anomaly response is a direct candidate when detection is followed by a bounded playbook. O-RAN testbeds have demonstrated radio-originating anomaly detection and mitigation through specialized xApps [28]. A System-One model could jointly estimate an incident class, severity level, and probability distribution over approved mitigation actions. Low confidence should lead to observation or escalation rather than an irreversible response. The same formulation applies to alarm prioritization and root-cause ranking, although free-form root-cause explanation is better delegated to an LLM or knowledge system. Other opportunities include slice admission accept,reject,defer\accept,reject,defer\, selection among deployment templates, traffic-steering activation, and cell or carrier sleep/wake control. These actions are bounded, but their effects are coupled across space and time. For example, switching off a lightly loaded cell may save energy while violating coverage or overloading a neighbor. A System-One model should therefore recommend a candidate management action and probability, while a digital twin, constraint checker, or optimizer validates its predicted impact. Natural-language intent interpretation is a less direct opportunity. User and operator intents may be ambiguous, cross-domain, and require clarification or decomposition; such tasks favor LLM-based agents. After an intent has been translated into explicit service objectives and admissible configurations, however, System-One models may support bounded choices such as policy selection, feasibility classification, conflict severity, and remediation routing. The boundary is therefore between open-ended intent understanding and typed operational decision making, not simply between management and control. TABLE I: Representative bounded decision opportunities across the wireless stack. Decision Representative state Admissible output Typical timescale Principal System-One value Link adaptation SINR/CQI, HARQ history, rank, mobility MCS or link-configuration index Slot to tens of ms Reliability-aware distribution and conservative fallback Beam selection Partial CSI, position, angles, past beams Beam or top-k beam subset Beam-coherence dependent Adaptive candidate-set size from confidence Scheduling/RRM Channels, queues, deadlines, slice KPIs Policy, user group, or candidate schedule Slot to near-RT Rapid choice among constrained profiles Handover Signal histories, velocity, load, QoS Stay or target cell Tens of ms to seconds Explicit ambiguity and safe defer/measure action Slice admission Resource state, SLA, predicted demand Accept, reject, defer, or template Seconds and above Typed decision with feasibility screening Anomaly response KPI/alarm history, topology, security events Monitor, isolate, throttle, rollback, escalate Near-RT to management Risk-aware mitigation and abstention Energy control Traffic forecast, coverage, neighbor load Active/sleep mode or policy Seconds to minutes Candidate policy selection before validation IV Experimental Evaluation We examine three complementary scenarios: receive-antenna selection, intent-conditioned RAN slicing, and intent-driven edge-service orchestration. The first two are implemented in our simulation framework, whereas the third draws on the independent study in [29]. Together, they cover numerical selection, joint semantic and numerical decision making, and integration with network-service execution. IV-A Evaluation Methodology Our evaluation compares Jev with frontier LLMs and task-specific classical baselines. Each method is evaluated on the same instances within each task, and its selected action is scored by a common numerical evaluator. Jev uses a categorical decision interface, while the LLMs are instructed to return a single action identifier in JSON format. Optimal actions and evaluator-generated scores are withheld from model inputs. The evaluation considers decision quality and response latency. For API-based methods, latency is measured at the client for the successful request, excluding deliberate pacing and retry waits. It therefore reflects the serving infrastructure and communication path as well as model execution. Classical methods are timed locally; these measurements provide implementation-level references rather than hardware-matched inference comparisons. IV-B Scenario I: Receive-Antenna Selection The first scenario considers a MIMO link with two transmit antennas and eight candidate receive antennas, of which two are selected to maximize instantaneous capacity. Ten independent Rayleigh-fading channel realizations are evaluated at SNRs of 0, 10, and 20 dB, yielding 30 channel–SNR instances and 28 admissible antenna pairs per instance. Exhaustive search provides the optimal reference. Practical baselines include norm-based selection, a correlation-based method (CBM), and random selection [30]. The AI methods receive the complex channel matrix, SNR, and candidate actions. Input variants examine the effect of capacity-oriented guidance and derived channel-strength and correlation features. For the LLMs, the guided variant includes both guidance and these features; Jev additionally separates guidance-only and feature-assisted inputs. Performance is assessed through relative capacity loss against exhaustive search, optimal-action selection rate, and decision latency. This scenario tests whether a general decision model can identify favorable antenna combinations from numerical channel information, and how much explicit wireless knowledge improves its decisions. Fig. 2: Mean relative capacity loss with respect to exhaustive search for receive-antenna selection. Fig. 3: Client-observed response latency of the AI methods for receive-antenna selection. Figs. 2 and 3 illustrate a tradeoff between decision quality and response latency. Jev has a client-observed latency of 1.52 s, compared with 12.86 s for GPT-Sol and 3.62 s for Qwen, corresponding to approximately 8.5×8.5× and 2.4×2.4× faster responses, respectively. However, its mean relative capacity loss is 20.19%, versus 0.00% for GPT-Sol and 17.33% for Qwen; the simple norm-based baseline also achieves a lower loss of 9.46%. Thus, in this mathematically demanding selection task, Jev’s faster response does not translate into better numerical decisions than the evaluated LLMs. These findings suggest that a low-latency, bounded decision interface alone is insufficient for accurate channel-dependent optimization. Given the early stage of System-One models, improving their numerical reasoning and wireless-specific grounding remains an important direction, potentially through integration with lightweight numerical solvers. Nevertheless, the present results cover only one task and should not be interpreted as a general limitation of the model class. IV-C Scenario I: Intent-Conditioned RAN Slicing The second scenario allocates 100 physical resource blocks among enhanced mobile broadband (eMBB), ultra-reliable low-latency communications (URLLC), and massive machine-type communications (mMTC). Allocations use increments of 20 resource blocks, producing 21 admissible actions. The benchmark comprises 30 instances formed from five service-intent profiles and six network-state variants. Each AI method receives a natural-language intent together with traffic demand, channel conditions, and queue information. A simplified numerical simulator scores the selected allocation using service satisfaction and penalties for applicable service-level agreement (SLA) violations. The underlying numerical intent weights and SLA annotations are reserved for evaluation. Exhaustive enumeration provides a full-information utility reference, while load-aware and random allocation provide practical baselines. The load-aware policy uses network state without interpreting the textual intent. The principal metrics are utility regret relative to the exhaustive reference, SLA violation severity, the fraction of allocations satisfying all applicable hard requirements, and decision latency. Utility and constraint satisfaction are assessed separately because the penalty-based utility optimum need not satisfy every hard requirement. Overloaded instances also allow examination of decisions when no admissible allocation satisfies all requirements. This scenario evaluates the ability to translate service intent into a bounded resource-allocation decision under competing demands. Fig. 4: Client-observed response latency of the AI methods for intent-conditioned RAN slicing. Fig. 5: Decision quality for intent-conditioned RAN slicing. Figs. 5 and 4 show a different behavior from the numerically structured antenna-selection experiment. In the intent-conditioned RAN slicing task, the learned decision models clearly outperform the random and load-aware baselines in terms of decision quality. In particular, Jev achieves a mean utility regret of 0.131, which is comparable to Qwen (0.134) and GPT-Sol (0.145), while maintaining substantially lower inference latency. As shown in Fig. 4, Jev reaches a median client-observed latency of only 1.46 s, compared with 5.13 s for Qwen and 5.29 s for GPT-Sol. These results suggest that, although current System-One models may still face limitations in tasks dominated by precise numerical computation, they can already provide competitive performance for compact intent-conditioned decision problems, where semantic requirements must be rapidly mapped to discrete control actions. Their favorable quality–latency tradeoff therefore makes them particularly promising for latency-sensitive wireless control as this emerging model family continues to mature. IV-D Scenario I: Intent-Driven Edge-Service Orchestration The third scenario draws on the independent evaluation by Li et al. [29]. Jev translates natural-language intents into service contracts specifying execution locations, deadlines, and priorities, while a numerical scheduler assigns tasks using current network and computing conditions. The experiments combine live model responses, packet-level New Radio simulation, mobility, shared edge queues, and an actual image-reading service. Fig. 6 compares decision latency across the cloud, local-model, and image-service experiments. Bars indicate medians, and whiskers show the 25th–75th percentiles rather than confidence intervals. Each experiment uses separate measurements, including its own Jev measurements; Qwen uses the direct-attribute interface throughout. In the cloud comparison, Jev achieves a median latency of 374.6 ms, versus 482.9 ms for DeepSeek and 984.5 ms for Gemini. Jev also provides lower median latency in the other two settings. All compared interpreters achieve correct interpretation on the evaluated contracts, with 63/63 checks in the cloud and local-model experiments and 48/48 in the image-service experiment. Fig. 7 presents the corresponding service outcomes. Cloud and local-model results average nine updated-contract trajectories, whereas image-service results measure correct, on-time completion among 1,080 requests per method. These different workloads require comparisons within each panel. In the cloud experiment, completion reaches 97.00% with Jev, compared with 93.51% for DeepSeek and 88.65% for Gemini. The local-model comparison likewise favors Jev, with 96.62% completion versus 91.84% for Qwen. However, in the image service, Jev completes 459 requests correctly and on time, compared with 463 for DeepSeek and 435 for Qwen. Faster interpretation therefore does not uniformly translate into higher end-to-end completion. These results illustrate the potential of combining bounded intent interpretation with numerical scheduling, while highlighting the need to assess application utility alongside decision latency. Both figures are redrawn from Tables 2–4 of [29]; these externally reported results remain separate from our simulations. Fig. 6: Decision latency for edge-service orchestration. Data from [29]. Fig. 7: Service completion for edge-service orchestration. Data from [29]. V Open Challenges and Research Directions The preceding opportunities depend on capabilities that are not yet established for System-One models in realistic wireless deployments. The main research questions concern transferable decision learning, reliable uncertainty, closed-loop safety, and deployability under network constraints. V-A Wireless-Native Decision Foundation Models An important direction is a wireless-native System-One foundation model, analogous in scope—but not in output mechanism—to foundation models in language. Existing wireless foundation models learn reusable representations from channel measurements, spectrograms, and heterogeneous wireless tasks [11, 12, 31, 14]. A decision-native model would instead be pretrained across many bounded control problems. For tasks τ∈τ , its training data could be written as =⋃τ∈(iτ,qτ,iτ,yiτ)i=1Nτ,D= _τ \(s^τ_i,q^τ,A^τ_i,y^τ_i)\_i=1^N_τ, (15) with the shared objective minθ∑τ∈wτ[ℓτ(pθ(⋅∣,q,),y)]. _θ _τ w_τE_D_τ\! [ _τ\! (p_θ(· ,q,A),y ) ]. (16) The goal is to reuse wireless structure across PHY, RRM, mobility, and management tasks while adapting to a new query or answer schema with little task-specific data. Major obstacles include incompatible modalities and timescales, inconsistent action semantics, scarce failure examples, negative transfer, and the absence of large, representative decision–outcome datasets. V-B Generalization and Calibrated Uncertainty Wireless distributions change with geography, frequency, hardware, mobility, traffic, and network configuration. Consequently, strong results on random train–test splits do not establish deployment robustness. Evaluation should include held-out cells, environments, devices, and decision schemas, together with out-of-distribution detection and risk–coverage analysis. Calibration must also be maintained after adaptation or continual learning; otherwise, the confidence used for abstention and escalation can become misleading precisely when the environment changes. V-C Constraint Satisfaction and Closed-Loop Safety A normalized decision distribution does not guarantee feasibility, stability, or acceptable long-term behavior. Hard power, coverage, protocol, and security constraints should therefore be enforced by deterministic checks or safety shields. Temporal effects also require explicit treatment: individually reasonable decisions may create ping-pong handovers, scheduler oscillations, or repeated sleep/wake transitions. Combining System-One inference with hysteresis, state history, optimizers, or digital twins is a promising route, but the resulting hybrid system must be verified as a whole. V-D Efficient Deployment and Reproducible Evaluation Practical performance is determined by end-to-end latency and energy, including measurement collection, state construction, inference, validation, signaling, and actuation. Compression, early exit, distributed inference, and privacy-preserving adaptation should be studied together with deadline-aware fallbacks. Progress also requires common datasets and benchmarks that compare System-One models with heuristics, optimization, and task-specific learning at matched compute and information budgets. In addition to average utility, reporting should include constraint violations, tail latency, calibration, selective risk, robustness under shift, and the frequency and cost of escalation. VI Conclusion This paper examined System-One models as a decision-native alternative to autoregressive LLMs for wireless control problems with explicit and bounded action spaces. Our evaluations show, however, that the value of this interface is strongly task dependent. In receive-antenna selection, Jev provided substantially lower response latency than the evaluated LLMs but could not match the decision quality of stronger numerical and LLM-based baselines. In contrast, for intent-conditioned RAN slicing, Jev achieved decision quality comparable to the evaluated LLMs while reducing response latency by more than threefold. Taken together, these results suggest that System-One models are most compelling as a complementary inference layer for repeated, bounded, and latency-sensitive decisions, particularly when semantic intent must be mapped rapidly to discrete control actions. A promising wireless architecture is therefore hybrid: System-One models can handle routine high-confidence decisions, while more capable reasoning models address uncertain, constrained, or computationally demanding cases. Realizing this vision will require wireless-native training, reliable calibration under distribution shift, explicit constraint enforcement, and reproducible end-to-end evaluation. As these capabilities mature, decision-native foundation models may provide a useful building block for scalable and responsive intelligence in future wireless networks. References [1] ITU-R (2023) Framework and overall objectives of the future development of IMT for 2030 and beyond. Technical report Technical Report Recommendation ITU-R M.2160-0, International Telecommunication Union. Cited by: §I. [2] AI-RAN Alliance (2024) AI-RAN alliance vision and mission white paper. Technical report AI-RAN Alliance. Cited by: §I. [3] T. J. O’Shea and J. Hoydis (2017) An introduction to deep learning for the physical layer. IEEE Transactions on Cognitive Communications and Networking 3 (4), p. 563–575. External Links: Document Cited by: §I. [4] H. Ye, G. Y. Li, and B. Juang (2018) Power of deep learning for channel estimation and signal detection in OFDM systems. IEEE Wireless Communications Letters 7 (1), p. 114–117. External Links: Document Cited by: §I. [5] S. Wang, H. Liu, P. H. Gomes, and B. Krishnamachari (2018) Deep reinforcement learning for dynamic multichannel access in wireless networks. IEEE Transactions on Cognitive Communications and Networking 4 (2), p. 257–265. External Links: Document Cited by: §I. [6] A. Feriani and E. Hossain (2021) Single and multi-agent deep reinforcement learning for AI-enabled wireless networks: a tutorial. IEEE Communications Surveys & Tutorials 23 (2), p. 1226–1252. External Links: Document Cited by: §I, §I-A. [7] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin (2017) Attention is all you need. In Advances in Neural Information Processing Systems, Vol. 30. Cited by: §I. [8] R. Bommasani, D. A. Hudson, E. Adeli, et al. (2021) On the opportunities and risks of foundation models. arXiv preprint arXiv:2108.07258. Cited by: §I. [9] H. Zou, Q. Zhao, Y. Tian, L. Bariah, F. Bader, T. Lestable, and M. Debbah (2024) TelecomGPT: a framework to build telecom-specific large language models. arXiv preprint arXiv:2407.09424. Cited by: §I. [10] D. Wu, X. Wang, Y. Qiao, Z. Wang, J. Jiang, S. Cui, and F. Wang (2024) NetLLM: adapting large language models for networking. arXiv preprint arXiv:2402.02338. Cited by: §I, §I. [11] A. Aboulfotouh, A. Eshaghbeigi, and H. Abou-Zeid (2025) Building 6G radio foundation models with transformer architectures. In Proceedings of the IEEE International Conference on Communications (ICC), p. 1888–1893. External Links: Document Cited by: §I, §V-A. [12] S. Alikhani, G. Charan, and A. Alkhateeb (2025) LWM: a pre-trained wireless foundation model for universal feature extraction. In Proceedings of the IEEE International Conference on Machine Learning for Communication and Networking (ICMLCN), External Links: Document Cited by: §I, §V-A. [13] H. Zou, Y. Tian, B. Wang, L. Bariah, S. Lasaulce, C. Huang, and M. Debbah (2026) RF-GPT: teaching AI to see the wireless world. arXiv preprint arXiv:2602.14833. Cited by: §I. [14] L. Liang, J. Guo, J. Zhang, C. Chae, L. Lu, S. Xu, O. A. Dobre, S. Jin, and G. Y. Li (2026) Foundation models for wireless communications: from PHY intelligence to network autonomy. arXiv preprint arXiv:2606.06239. Cited by: §I, §V-A. [15] TypeSafe AI (2026) Introducing system one models and Jev. Note: Technical announcementVendor-provided description and evaluation; accessed September 23, 2026 Cited by: §I, §I, §I. [16] N. M. and ConvAI Innovations (2026) Laya: multilingual, non-autoregressive system one decision model. Note: Open-weight model and GitHub repositoryDeveloper-provided model card and benchmarks; accessed September 23, 2026 Cited by: §I, §I. [17] J. Zhang, T. Yang, Y. Shi, and L. Wu (2026) AnyJev: turn any LLM into a Jev-style decision model. Note: Open-source software repositoryAccessed September 30, 2026 External Links: Link Cited by: §I. [18] TypeSafe AI (2026) Deel evaluation of Jev on production decision tasks. Note: LinkedIn post reporting customer-side evaluation resultsResults attributed to Deel; accessed September 30, 2026 External Links: Link Cited by: Fig. 1, §I. [19] O-RAN Alliance Next Generation Research Group (2024) dApp use cases and requirements. Technical report Technical Report nGRG-R-2024-10, Version 2.21, O-RAN Alliance. Cited by: §I-A. [20] H. Lee, D. Kim, and K. Min (2025) Self-improving interference management based on deep learning with uncertainty quantification. IEEE Transactions on Vehicular Technology 74 (12), p. 19927–19932. External Links: Document Cited by: §I-A. [21] J. Zhang, J. Li, L. Shi, Z. Wang, S. Jin, W. Chen, and H. V. Poor (2025) Decision transformers for wireless communications: a new paradigm of resource management. IEEE Wireless Communications 32 (2), p. 180–186. External Links: Document Cited by: §I-A. [22] R. Sun, N. Cheng, C. Li, W. Quan, H. Zhou, Y. Wang, W. Zhang, and X. Shen (2025) A comprehensive survey of knowledge-driven deep learning for intelligent wireless network optimization in 6G. IEEE Communications Surveys & Tutorials. External Links: Document Cited by: §I-A. [23] 3GPP (2024) NR; physical layer procedures for data. Technical report Technical Report TS 38.214, Version 18.4.0, 3rd Generation Partnership Project. Cited by: §I-B. [24] F. J. Martín-Vega, J. C. Ruiz-Sicilia, M. C. Aguayo-Torres, and G. Gómez (2021) Emerging tools for link adaptation on 5G NR and beyond: challenges and opportunities. IEEE Access 9, p. 126976–126987. External Links: Document Cited by: §I-B. [25] C. Shao, C. Liu, L. Zhao, M. Li, X. Zhang, and M. Sun (2025) Deep learning-based millimeter wave beam recommendation via channel knowledge map. IEEE Wireless Communications Letters 14 (6), p. 1648–1652. External Links: Document Cited by: §I-B. [26] C. V. Nahum, V. H. L. Lopes, R. M. Dreifuerst, P. Batista, I. Correa, K. V. Cardoso, A. Klautau, and R. W. Heath (2024) Intent-aware radio resource scheduling in a RAN slicing scenario using reinforcement learning. IEEE Transactions on Wireless Communications 23 (3), p. 2253–2267. External Links: Document Cited by: §I-C. [27] A. Lacava, M. Polese, R. Sivaraj, R. Soundrarajan, B. S. Bhati, T. Singh, T. Zugno, F. Cuomo, and T. Melodia (2024) Programmable and customized intelligence for traffic steering in 5G networks using open RAN architectures. IEEE Transactions on Mobile Computing 23 (4), p. 2882–2897. External Links: Document Cited by: §I-C. [28] H. Bogucka, M. Hoffmann, P. Kryszkiewicz, and Ł. Kułacz (2025) An open-RAN testbed for detecting and mitigating radio-access anomalies. IEEE Communications Magazine 63 (11), p. 122–127. External Links: Document Cited by: §I-D. [29] D. Li, X. Wang, H. Gong, R. Lang, and G. Yu (2026) Fast intent-driven service orchestration with Jev for 6G edge networks. arXiv preprint arXiv:2609.23136. Cited by: Fig. 6, Fig. 7, §IV-D, §IV-D, §IV. [30] A. F. Molisch, M. Z. Win, Y. Choi, and J. H. Winters (2005) Capacity of mimo systems with antenna selection. IEEE Transactions on Wireless Communications 4 (4), p. 1759–1772. Cited by: §IV-B. [31] T. Zheng, J. Guo, L. Dai, S. Jin, and J. Zhang (2026) MUSE-FM: multi-task environment-aware foundation model for wireless communications. IEEE Transactions on Wireless Communications 25, p. 19791–19806. External Links: Document Cited by: §V-A.