Language Model Reasoning Evaluation

Follow this topic Ask about this topic

Latest papers

128 papers66 papers in the last four weeks, up 2100% on the four weeks before. 0.4% of all new papers.

Oct 19, 2026

Diagnostic Analysis for Large Language Model Reasoning

Reasoning structures (e.g., chains and graphs) are increasingly used for probabilistic inference, factuality verification, and uncertainty estimation, but their structural reliability is rarely evaluated directly. We test whether this reliability holds through a proposed two-stage diagnostic evaluation in open-ended multi-hop QA. Stage 1 tests different capabilities required to construct reasoning graphs, showing that graph reliability varies with topology, prompt design, and model family. Stage 2 uses uncertainty estimation as a downstream test case. Generated-edge experiments show that graph-based computation is sensitive to upstream graph errors, with performance dropping relative to the gold-graph setting. Gold-graph experiments further show that graph structure provides little benefit when used only as prompt context, but improves the separation of correct from incorrect answers by confidence score and enables node-level diagnostics.

Xinyu Liu, Dongqi Fu, Jun Wu

Michigan State University · Meta AI

Language Model Reasoning Evaluation LLM Reasoning with Graphs Multi-Hop QA

Oct 19, 2026

Answer Delivery Under a Shared Token Budget: An Audited Study of Stopping and Finalization

An inference budget is useful only when the system delivers an answer within it. We study a fixed-budget policy combining first-valid-box stopping with a charged finalization cue. On 100 fresh, independently audited GSM8K test questions per model, it improves delivered endpoint accuracy over first-box stopping by 8.0 percentage points for Qwen3-4B-Thinking-2507 and 7.0 for DeepSeek-R1-Distill-Llama-8B. A prelocked repeat on the same questions gives 7.0 and 10.0 points; neither seed shows a gross harm relative to first-box stopping. The point estimates of incremental physical-slot savings are below one percent. A finalization-only ablation separates cue recovery from stopping, and a fixed scalar reader tests extraction as an alternative explanation. The cue is prior art; the contribution is a component and readout analysis with charged injections and verified paths. Results cover two checkpoints, one task, one prompt family, and two fixed seed conditions.

Yupeng Chang, Linyuan Gao, Zhiyuan Li +2

School of Artificial Intelligence, Jilin University · Engineering Research Center of Knowledge-Driven Human-Machine Intelligence, MOE, China · International Center of Future Science, Jilin University

Efficient Language Model Inference Early Stopping Inference-Time Guidance Language Model Reasoning Evaluation

Oct 19, 2026

Understanding When Reflective Prompt and Skill Optimization Helps on Locally Served Reasoning Models: A Two-Model Study of GEPA and SkillOpt

Reflective text-space optimizers such as GEPA and SkillOpt have reported large gains on frontier and mid-size models. We test replication across two reasoning-native target-serving configurations: DeepSeek V4 Flash through DwarfStar and Qwen3.8-27B through LM Studio. For each target, we evaluate the same held-out tasks and four conditions - minimal baseline, GEPA prompt, SkillOpt skill, and manually engineered prompt - on HotpotQA (n=500 per condition) and MATH-500 (n=250). In the primary Qwen v2 rerun under the same frozen protocol, GEPA selects the HotpotQA seed and exactly reproduces baseline behavior, whereas SkillOpt's validation gain reverses into held-out degradation. On MATH-500, GEPA raises exact match from 0.828 to 0.904 while reducing latency but increasing token consumption, so lower response time does not imply lower token cost. DeepSeek likewise shows its strongest GEPA effect on MATH-500 (0.852 to 0.908). Across both targets, GEPA's strongest gains occur on MATH-500, while neither optimizer improves the HotpotQA baseline. Optimization is therefore task- and method-dependent, and validation gains may reverse on held-out evaluation. An earlier Qwen run is retained as serving-sensitivity evidence. These results suggest that textual headroom for a model-task pair and the observed effect of a text-space optimizer are more informative when decomposed into content/reasoning, answer-realization/evaluator alignment, and serving-completion/reliability. The cross-benchmark pattern further suggests that optimization may benefit from low-ambiguity, actionable evaluation signals.

Andrea Giummarra, Carmelo Baglieri, Antonio Di Maria +3

Evision srl, Ragusa, Italy · Department of Clinical and Experimental Medicine, University of Catania, Catania, Italy

LLM Input Optimization Large Language Model-Guided Optimization LLM Reasoning Language Model Reasoning Evaluation

Oct 19, 2026

ABACUS: Program-Guided Multistep Reasoning over Frozen Judgment Models

Judgment models answer semantic questions by scoring a finite set of candidate outcomes. Can reusable programs turn these local decisions into powerful multistep reasoning? We introduce \method, a framework that jointly constructs judgment schemas and composition programs for frozen judgment models. A generative teacher designs the procedure offline; the judgment model resolves local semantic questions online, and the saved program composes its decisions into an answer. We study three allocations of design effort between humans and the teacher, focusing on construction through a shared interface. We instantiate the framework with Kev. On three selected BIG-Bench Hard tasks, nine independently constructed programs solve all 450 executions of 150 development inputs. A separate deployment study freezes three programs after development-set reference-agreement refinement and evaluates them on 600 held-out questions. These programs achieve near-perfect 99.67% macro accuracy, versus 64.00% for direct Kev and 68.80% for final-answer-only Qwen3.5-4B. They attain comparable observed accuracy to Qwen's unrestricted response policy (99.07%) with an 8.43-fold online speedup: 0.255 versus 2.150 seconds per question. These results show how reusable programs can substantially expand the reasoning capability of a compact judgment model while keeping its parameters fixed.

Yuhao Wang, Zelin Zang

Westlake University · Hong Kong Institute of Science & Innovation, CAS · Tsientang Institute for Advanced Study

Structured Reasoning in Language Models Efficient Language Model Reasoning LLM-Based Program Synthesis Language Model Reasoning Evaluation

Oct 19, 2026

Truncated Reasoning Under a Budget: Evaluating System-1 Models as Decision-Makers

Large language models (LLMs) often generate more reasoning tokens than needed. Limiting the reasoning budget reduces cost, but forcing an answer from an incomplete trace can hurt accuracy. System-1 decision models select among candidate options without producing a reasoning response. We evaluate Jev as a decision reader of truncated reasoning across seven LLM reasoners and five multiple-choice benchmarks (35 pairs). Jev achieves higher mean accuracy than forced answering at tested budgets from 500 to 16,000 tokens, including the estimated costs of reasoning and answering. At a 1,500-token budget, it improves accuracy by 4.1 pp overall and 8.6 pp on truncated questions, with significant gains in 17 pairs under Holm-corrected bootstrap tests. Offline replay estimates 34.5% of the full-run cost, with a 1.13 pp accuracy loss. A policy selected on training folds saves 71.3% of the estimated full-run cost on held-out folds, with a 0.58 pp accuracy loss. We also find that Jev shows strong question-only accuracy, exceeding 500-token forced answering on average (68.0% vs. 65.5%), although this advantage varies by benchmark. These findings suggest that System-1 decision models can complement LLM reasoning and reduce the need for costly reasoning completion.

Zhongyin Wang, Zhaoqian Xue, Junting Mu +4

University of Illinois Urbana-Champaign · University of Arizona · University of California, Irvine +3

LLM Decision-Making Efficient Language Model Reasoning Language Model Reasoning Evaluation

Oct 19, 2026

Which Answers Agree? Evaluating Self-Consistency from Incomplete Logs

Self-consistency answers a question by generating multiple responses and selecting the most frequent answer. Its evaluation requires both correctness labels and information about which responses agree; correctness labels alone generally determine neither voting accuracy nor the benefit of additional votes. For tasks in which all correct responses map to the same normalized answer, we derive sharp accuracy bounds using classical stochastic majorization, refined by abstention counts and answer-category limits. We also derive sharp bounds on accuracy changes across voting budgets that share a response pool. When answer-category counts are available, we introduce Plurality-U, a count-based dynamic program that exactly averages voting accuracy over all fixed-size subsets, eliminating finite-pool Monte Carlo error. On 1,403,520 stored outputs from recent reasoning models in Scorio Lite and Scorio Trace, recording abstention counts and answer options cuts the width of SuperGPQA accuracy intervals by more than half. For moving from 9 to 17 votes, our bounds reduce interval width by a median of 77.7% on SuperGPQA and 72.3% on 14 GPQA and Adaptive-Consistency pools, compared with a baseline that combines separate accuracy bounds. Exact computation has lower paired median runtime than 2,000-subset Monte Carlo in 14 of 15 tested pool-budget settings. These results clarify what incomplete logs can establish and provide an exact evaluation procedure when answer counts are retained.

Zhengtao Gui, Zexi Chen, Weining Shen

University of California, Irvine

Self-Consistency in Language Models Language Model Reasoning Evaluation

Oct 18, 2026

Assertion and Belief: Moore's Paradox in Language Models

Moore's paradox \citep{moore1942} concerns sentences such as ``It is raining, but I don't believe that it is,'' which may well be true yet cannot be sensibly asserted. The absurdity lies not in the content but in the act, since to assert is to vouch for one's own belief. This makes the paradox a test of whether language models take the speech act into account when computing meaning. We present the first test of Moore's paradox in LLMs, evaluating 12 models on a curated set of Moorean sentences in varied contexts. We find that models largely recognize the infelicity of Moorean sentences, and that the judgement is not tied to a memorized form and tracks the speech act. Using the Rational Speech Act framework (RSA; \citealt{goodmanfrank2016}), we then test whether these judgements arise from a listener reasoning about a speaker. We find the models' own listeners do not reason about their own speakers, and their reasoning traces suggest that they instead recall knowledge about the paradox and apply rules therein. The models thus appear to have learned the paradox as a fact about sentences, not as a requirement on speakers.

Anshun Asher Zheng, Weili Cao, Fangcong Yin +2

The University of Texas at Austin · Duke University · New York University +1

Pragmatic Reasoning in Language Models Language Model Reasoning Evaluation

Oct 18, 2026

Stepwise Reasoning Verification: Don't Generate, Classify!

Verifying the reasoning of large language models is increasingly important as they are applied to complex mathematical and scientific problems. Existing work focuses on large decoder models and reports mixed results for generative and discriminative verifiers, leaving unclear which paradigm is more effective and whether large models are necessary. In this work, we cast stepwise verification as a natural language inference task: the problem and partial solution prefix form the premise, while the next reasoning step serves as the hypothesis to be classified as entailed or non-entailed. This formulation enables error localisation at individual reasoning steps and direct comparison of encoder classification, decoder classification, and decoder verdict generation across model scales and annotation budgets. Across five reasoning datasets, a 0.15B encoder classifier outperforms most generative verifiers with up to 4B parameters. On identical decoder backbones, classification outperforms generation from 0.5B to 4B parameters and matches it on the strongest 7B backbone---reaching comparable performance with roughly one tenth as many annotated problems. Our findings establish classification as a parameter- and data-efficient approach to stepwise verification and show that strong verification does not require large generative decoders.

Yifei Shi, Michalis Korakakis, Rami Aly +3

University of Cambridge · Alan Turing Institute · University of Edinburgh

Reasoning Verification Language Model Reasoning Evaluation Decoder-Only Language Models

Oct 18, 2026

Where and Why Search-Augmented LLMs Fail When Reasoning over Rapidly Evolving Knowledge

As the world changes, new information is published while older records remain online. This complicates reasoning for large language models (LLMs), which must retrieve evidence and resolve conflicts to determine which information remains valid. Existing evaluations treat information seeking as a black box, revealing little about where and why LLMs fail. To address this gap, we introduce TREK, a suite of three components: TREKTRACE is a plug-and-play framework that attributes failures to specific stages of retrieval and reasoning. We apply it to TREKBENCH-WILD, a meta-benchmark of real-world questions that require complex reasoning over rapidly evolving knowledge. Surprisingly, the main bottleneck is post-retrieval reasoning rather than retrieval: frontier models are often misled by conflicting evidence even after retrieving all the information needed to answer. To study this bottleneck, we build TREKBENCH-CTRL, a controlled benchmark with distractors spanning temporality, relevance, and trustworthiness, and show that benchmarks dominated by outdated distractors substantially underestimate task difficulty. Together, TREK provides a testbed for diagnosing and improving reasoning over rapidly evolving knowledge.

Tianyi Hu, Bhavyajeet Singh, Niket Tandon +1

Aarhus University · Microsoft Research · Technische Universität Darmstadt

Retrieval-Augmented Reasoning LLM Reasoning Language Model Reasoning Evaluation Knowledge Conflicts in Language Models Temporal Reasoning in Language Models

Oct 18, 2026

MAMath-Bench: Process-Aware and Robust Evaluation of Multimodal Multi-Agent Mathematical Reasoning

Mathematical reasoning is a fundamental capability for educational applications of Large Language Models (LLMs), where reliable problem solving requires not only correct answers but also coherent and verifiable reasoning processes. Multi-agent systems (MASs) have shown growing potential in this domain by leveraging collaboration and verification among multiple LLM agents. However, existing evaluations primarily focus on the correctness of final answers in text-only settings, offering limited insight into multimodal reasoning capabilities, the reliability of intermediate reasoning steps, and robustness against reasoning errors. To address these limitations, we introduce MAMath-Bench, a process-aware benchmark designed to evaluate the multimodal mathematical reasoning capabilities of MASs. MAMath-Bench comprises 1,003 problems spanning 11 mathematical categories, multiple difficulty levels, and several modalities, including text, images, and audio. It features fine-grained milestone annotations to evaluate systems across three dimensions: problem-solving correctness, reasoning-process reliability, and robustness to adversarial reasoning-chain perturbations. Our evaluation yields three key findings. First, multi-agent collaboration does not consistently improve final-answer accuracy over strong single LLMs. Second, collaboration can substantially enhance the coverage of intermediate reasoning, but such gains may come at the cost of increased reasoning redundancy. Third, current systems remain vulnerable to adversarial reasoning errors, particularly those involving subtle structural perturbations. The dataset is available at https://github.com/mira-ai-lab/AgentMath.

Xinping Wang, Jie Ma, Rui Xing +6

School of Cyber Science and Engineering, Xi’an Jiaotong University · MOE KLINNS Lab, Xi’an Jiaotong University · School of Computer Science and Technology, Xi’an Jiaotong University +3

Mathematical Reasoning Benchmarks Language Model Reasoning Evaluation Multi-Agent Reasoning Multimodal Reasoning

Oct 18, 2026

What Can't Jev Decide? Probing the Limits of One-Pass Decision Models

Jev and its open replicas choose among the options a caller supplies with a single forward pass, fast enough to decide inside an agent loop. We ask what such a model cannot decide, and find a simple limit: Jev cannot make a decision that an LLM makes only by writing out intermediate steps. On a benchmark that varies one factor at a time, Jev finds and selects records without error up to 20k tokens and 64 options, but once the answer depends on two or more updates of one user, accuracy falls, to 53-80% at sixteen updates, because similar records interfere. Given a stale summary of the state, Jev follows it. On real decision items graded by how many thinking tokens a reference LLM needs, Jev answers 97.5% of the items that one forward pass of that LLM solves, but only 74% of the others. Jev also does not know its limit: its confidence flags the errors it makes itself but not those it inherits from a stale input, and neither an explicit nor a trained hand-off option catches the latter. The open models we evaluate share these limits. Checking the state such models rely on requires a model that reasons.

Wang Yang, Xinpeng Li, Yanyan Zhang +8

Case Western Reserve University · NII LLMC (Japan) · Georgia Institute of Technology +1

State Tracking in Language Models LLM Decision-Making LLM Reliability Language Model Reasoning Evaluation

Oct 17, 2026

Verifying Human-Written Reasoning References

Evaluating the reasoning of AI systems often relies on human-written reasoning chains as references, yet the logical support for their individual steps is rarely verified. Although prior neuro-symbolic approaches have formalized and checked reasoning, their reported evaluations do not systematically assess whether step conclusions in human-written reference derivations are entailed by their cited premise sets or investigate the sources of failures in those references. To address these limitations, we present a novel reference-centred assessment method for human-written derivations that combines entailment checking against cited premise sets, full-premise rechecking, and failure attribution to examine the reliability of these derivations and their formal representations. We evaluate it on P-FOLIO, using a large language model (LLM) to translate step conclusions into first-order logic (FOL) and the Z3 SMT solver to check entailment. Of 386 derivations examined, 202 (52.3%) meet all our checking criteria, meaning that every translated step is entailed by its stated support and the required target is established under the available FOL representations. Among 230 steps not entailed by their stated support, 191 become entailed when checked against the complete original premises, indicating that most initial failures reflect incomplete or mismatched support rather than unsupported claims. Manual analysis of the remaining 39 cases identifies 21 human reasoning errors, 11 FOL translation errors, and 7 dataset annotation errors. A separate manual audit of 50 randomly sampled steps that passed the entailment checks found 46 faithful FOL translations (92%). These findings suggest that disagreement with a human-written reasoning reference may originate in the reference or its formal representation rather than in the AI system being evaluated.

Ali Ansari, Farhana Ferdousi Liza

School of Computing Sciences, University of East Anglia

Reasoning Verification Language Model Reasoning Evaluation

Oct 17, 2026

What RL Brings to Reasoning: A Few Decisions the Base Model Would Not Make

Reinforcement learning with verifiable rewards (RLVR) substantially improves mathematical reasoning, but what it adds to the base model remains unclear. We investigate this through a sequence of generation-time interventions on Qwen3-4B-Base, measuring how much of its RL-trained counterpart's gain each intervention recovers on 120 AIME problems. Power sampling with the same 8,192-token output limit brings no gain beyond greedy decoding. Yet borrowing the RL model's token only where the two models' greedy choices differ, roughly 4% of generated positions, recovers 90.5% of the pass@1 gain, with similar recovery across three model pairs on MATH-500. We find that even a single opening token has a substantial effect: forcing the base model to begin with "Alright" recovers 35% of the RL gain. The opening alone is insufficient, however. Later replacements after the same opening add another 5.2 percentage points to pass@1, and the fixed rules we test for inserting reflective words such as "Wait" do not reproduce this benefit. Sparse replacement also retains much of the base model's diversity while staying close to the RL model at pass@1. At k=256, it solves 67.5% of the problems, significantly more than the RL model (61.4%) and a mixture of whole responses matched in pass@1 (61.45%). The comparison is at equal attempt counts rather than equal compute, since rl-replace consults both models at every decoding step. These results show that the opening can recover a meaningful part of the RL gain, but a larger contribution comes from later choices, which can be borrowed while retaining much of the base model's diversity. Code is available at https://github.com/CyrusZhang04/WhatRLBrings.

Siyu Zhang, Yuquan Lu, Yoshua Y. Li +7

University of California, San Diego · Meituan, LongCat Team · The Chinese University of Hong Kong, Shenzhen

RL for Language Model Reasoning LLM Reasoning Language Model Steering Language Model Reasoning Evaluation

Oct 17, 2026

CoTs remain monitorable when being unfaithful requires computation: Replicating and Extending the Necessity Effect Across Eleven Models

While a LLM's chain of thought (CoT) is a promising approach to interpreting model outputs, prior work shows CoT's content does not always accurately reflect the reasoning process. Emmons et al. (2025) finds that models can follow a question hint without revealing so in its chain of thought, but when the hint is sufficiently complex the computation is forced into the CoT. This claim, however, has only been tested on Gemini models, and the datasets and code is closed source. We replicate their proxy task across eleven models from six families and find the necessity effect generalizes. Every model follows simple cues above a no-hint baseline, while concealed complex-cue following falls back near baseline for most models. We also extend the study from Emmons et al. (2025) to measure CoT-necessity and unfaithfulness across five cue families (four of them new) and three difficulty levels. The relationship between when CoT is necessary and unfaithfulness is task-specific and model-specific. For some hint types, models often verbalize their hint-following behavior even when the hint can be decoded in a single forward pass. These results support the case for CoT monitoring when unsafe behavior requires non-trivial computation, and that properties of CoT unfaithfulness are sensitive to model and setting.

Arav Dhoot, Yixiong Hao, Zephaniah Roe

Columbia University · Second Look Research

Faithfulness of Language Model Reasoning Faithfulness of Language Model Explanations CoT Monitoring Language Model Reasoning Evaluation

Oct 16, 2026

Distillation, Self-Training, and the Fragility of Learned Chemical Reasoning

Automating synthesis planning for small-molecule drug design requires chemical reasoners that are accurate as well as open to inspection and adaptation, yet frontier language models remain unreliable at predicting reaction outcomes and difficult to deploy, while small open-weight models lack the necessary domain reasoning. We show that frontier supervision can serve as a one-time bootstrap for open models on forward reaction prediction. Our models, Resonance, distill an open-weight teacher's reasoning traces and then learn from their own verified outputs, turning a small base model with near-zero accuracy into a capable reaction predictor, while a stronger student surpasses its teacher through distillation alone. Accuracy, however, does not bring robust reasoning: the small student inherits its teacher's single-pass style, self-training mainly sharpens answers it could already reach, and it readily abandons correct answers under pushback, suggesting that chemical reasoning models should be evaluated on the robustness of their reasoning as well as on the accuracy of their answers.

Anagha Aneesh, Sebastian Ibarraran, Frank Hu +2

Stanford University

Chemical Reasoning in LLMs Chemical Reaction Prediction Language Model Distillation Language Model Robustness Self-Training Language Model Reasoning Evaluation

Oct 16, 2026

Reasoning Models Can Control Their Chains of Thought When They Tell Themselves To

Chain-of-thought (CoT) controllability, the ability of reasoning models to control properties of their CoT, is consequential for both their usefulness and reliable oversight. Recent work evaluates this capability by measuring whether models follow CoT constraints given in the user prompt, and reports very low controllability across models. We argue that this low controllability may not reflect an inherent inability of reasoning models to control their CoT, but instead result from naively specifying constraints in the user prompt. To test this, we introduce Self-Instructed CoT Controllability: a complementary evaluation that places the constraint as a self-instruction inside the model's CoT and measures its ability to follow it. We show that, under self-instruction, models achieve significantly higher controllability than with standard user instructions. Even a generic reminder to follow the user's constraint substantially improves controllability. Moreover, effective self-instructions can be bootstrapped into the reasoning through a simple instruction in the user prompt, enabling improved control even without access to the reasoning channel. Finally, using self-instruction, we demonstrate in a realistic monitoring setup that CoT controllability can be leveraged for monitor evasion, posing a risk to CoT monitorability.

Uriel Dolev, Idan Schwartz, Yoav Goldberg

Bar-Ilan University · Allen Institute for AI

Language Model Reasoning Evaluation LLM Prompting CoT Monitoring LLM Safety Monitoring

Oct 16, 2026

Not All Test-Time Compute Is Evidence: Model-Specific Bias in LLM Reasoning

Test-time scaling methods implicitly assume that additional computation from the same model provides independent evidence. We show that this assumption fails within a single reasoning model: errors are correlated across repeated samples, across reasoning depth, and in the model's own self-evaluation, so more compute from the same model is not more evidence. We introduce DIA (Disagreement-triggered Independent Arbitration), which treats disagreement between an intermediate-stage and the final-stage prediction as a per-question signal of reasoning instability and resolves it by asking a small panel of heterogeneous models to solve the question independently, without seeing either candidate, and aggregating their answers directly. Across seven benchmarks and two base models, DIA outperforms full chain-of-thought in 12 of 14 settings, including a 7.00% gain on MuSiQue, while activating on only

\sim 17\%

of examples and using up to 47.2% fewer tokens than a multi-agent baseline. Ablations show that both independent solving and solver heterogeneity are necessary for these gains.

Ming Shen, Tianjun Yao

Shanghai Jiao Tong University · Shenzhen Loop Area Institute (SLAI)

Test-Time Scaling Efficient Language Model Reasoning Reasoning Consistency in Language Models Language Model Reasoning Evaluation

Oct 16, 2026

Do Knowledge and Reasoning Oriented Benchmarks Induce Stable Representation Signatures in LLMs?

Internal signatures could support monitoring, adaptive computation, and mechanistic analysis, but benchmark prompts can be separable without any shared computational distinction. We test whether a prespecified knowledge/reasoning label defines a reusable internal signature in five contemporary 7B to 31B models. We treat the task as the unit of inference probes hold out complete tasks, labels are permuted across tasks, and the best layer and stream are reselected for every permutation. Raw probes recover an eight-versus-eight-subject MMLU partition in all five models (macro-F1

0.78

--

0.81

), but prompt text predicts it nearly as well: under an identical probe configuration, the activation advantage is only

0.017

--

0.070

macro-F1 and is not significant. At the prespecified

0.01

level, lexical adjustment removes significance while the adjusted states still identify individual subjects, and the selected site changes across task panels and selection rules. Additionally, two released base-training trajectories show no common direction. These pooled layerwise measures recover the task label but identify no transferable knowledge/reasoning signature at the tested sensitivity, which cannot exclude moderate effects. The resulting protocol gives future internal monitors and intervention studies a sequence of tests to pass. Project page: https://parsa-ra.github.io/lmbehave/.

Parsa Rahimi, Azim Dehghani Amirabad

EPFL, Idiap Research Institute · Independent Researcher

LLM Interpretability Language Model Probing Language Model Reasoning Evaluation

Oct 15, 2026

Are Large Reasoning Models Good Everyday Reasoners? A Comparison with Non-Reasoning Models on Logical Reasoning Tasks

Large reasoning models (LRMs) that leverage test-time computation achieve strong performance on mathematical and scientific benchmarks and are increasingly deployed in real-world applications. However, their ability to perform everyday logical reasoning, which often departs from textbook formal logic, remains underexplored. In this work, we systematically evaluate how well reasoning models align with everyday reasoning patterns by introducing a benchmark based on linguistic and pragmatic reasoning phenomena. We also compare reasoning models with non-reasoning models under controlled prompting conditions. We find that, in many cases, (1) reasoning models show degraded performance in everyday reasoning when such reasoning deviates from formal logical rules, (2) some non-reasoning models with chain-of-thought prompting outperform comparable reasoning models in everyday reasoning, and (3) prompting with additional instructions provides limited control over reasoning behavior. These findings suggest that current reasoning models do not reliably capture everyday reasoning and point to limitations in post-training, motivating improved training objectives and evaluation across diverse forms of reasoning.

Kentaro Ozeki, Risako Ando, Takanobu Morishita +3

Keio University

Language Model Reasoning Evaluation LLM Reasoning Large Reasoning Models Pragmatic Reasoning in Language Models

Oct 14, 2026

RCCM: Diagnosing Gains and Harms in Reasoning-Prompt Transfer

A reasoning prompt can improve one deployment and degrade another. RCCM (Reasoning-Control Capability Mismatch) examines these reversals by pairing answers to the same questions with records of how generation ends. Across five open-weight deployments and nine strategies, aggregate accuracy hides substantial question-level gains and harms; a common-budget MATH500 evaluation examines transfer on a separate sample. On 60 additional questions, concise reasoning improves Phi's automatic answer matching at a 2,048-token cap across three collections, but sharply reduces Qwen's. Author-reviewed final-answer commitments give a smaller, statistically unresolved Phi gain. Reserving an answer stage also improves matching, with additional input and call costs. These results show why a prompt benefit must be interpreted alongside its execution and evaluation: matching a reference answer need not mean delivering it, and an apparent compression benefit may admit another explanation. RCCM records this evidence for reproducible transfer comparisons.

Yuqi Zhang, Di Zhang

Language Model Reasoning Evaluation LLM Reasoning Language Model Generation Evaluation Efficient Language Model Reasoning

Sep 24, 2026

Reasoning Instructions Can Break Answer Decoding in Vision--Language Models

Chain-of-thought (CoT) instructions can distort multiple-choice VLM evaluation when a scorer appends a reasoning cue but reads answer-label logits before the model generates any rationale. We call this CoT-prefix scoring. On ScienceQA, Qwen2.5-VL-7B drops from 80.76% to 45.48%, and across five option-content permutations 93.54% of CoT-prefix predictions select the first slot. Condition-matched linear probes recover 78.94% from the same hidden states, while free generation restores 75.24%, showing that the answer often survives the prefix and the immediate readout fails. Vocabulary and layer diagnostics explain the mismatch: probability mass moves toward continuation tokens, while answer information remains linearly accessible in late layers. The effect recurs with varying severity across datasets and models, though not universally. These results show that CoT-prefix scoring can confound model knowledge with an evaluation-interface mismatch and should be avoided unless the requested and scored output events are aligned.

Zeyan Li, Siyuan Qiu, Jianfeng Xu

Shanghai Jiao Tong University

VLM Evaluation Language Model Generation Evaluation Multiple-Choice Question Answering CoT Reasoning Language Model Reasoning Evaluation

Sep 19, 2026

LLaDA-PRM: A Bidirectional Step-Level Reasoning Evaluator

Step-level reasoning evaluators are commonly based on autoregressive language models, whose causal attention restricts each step representation to the problem, previous steps, and the current step. Yet, when the complete solution is available, the validity of an earlier step may become clearer only through its downstream consequences. We validate this hypothesis through a controlled 54-run comparison of causal and bidirectional LLaDA evaluators at 1B--3B scale, changing only the self-attention mask, and find bidirectional attention yields consistent improvements. Building on this finding, we introduce \prm{}, an 8B bidirectional evaluator that reaches 88.8 step-level F1 on MR-MATH-invalid and 83.8 on the out-of-distribution MR-GSM8K original-question subset, outperforming ReasonEval-Llemma-34B by 11.3 and 10.3 F1 points, respectively. \prm{} also remains effective when evaluating incomplete reasoning traces in online settings, outperforming the strongest baselines on both benchmarks by a large margin. We further show that \prm{} provides an effective training-data selection signal, improving Mistral-7B performance on MATH-500.

Yiming Feng, Naihao Deng, Yulong Chen +1

University of Michigan · University of Aberdeen · University of Cambridge

Reasoning Evaluation Language Model Reasoning Evaluation Process Reward Models LLM Evaluation Language Modeling Numerical Reasoning in Language Models

Sep 8, 2026

Do Reasoning Representations Help Humans Evaluate LLM Outputs?

Reasoning representations are increasingly used as explanations for large language model outputs. Yet they are typically evaluated with model-centric criteria, such as answer accuracy and faithfulness, leaving it unclear whether they help people evaluate model responses. In this work, we study reasoning representations as human-facing interfaces rather than proxies for model reasoning ability. We conduct a controlled human study of six reasoning formats across tasks of varying complexity, supported by a web-based framework that randomizes task domains, problem instances, and representation order. The study collects fine-grained judgments of structural understanding, error detection and localization, and trust calibration. Our study shows a mismatch between perceived preference and support for human evaluation. Participants prefer planning- and decomposition-based representations, but simpler chain-of-thought traces better support verification, trust, and interpretability. Preferred representations also introduce calibration risks, with more false alarms on correct traces and high trust despite low willingness to verify.

Jaewoo Lim, Sungbok Shin, Sanghyun Hong

Oregon State University, Corvallis USA · Sogang University, Seoul, South Korea

LLM Evaluation LLM Interpretability CoT Reasoning Human-in-the-Loop Evaluation Explainability Evaluation Language Model Reasoning Evaluation

Aug 31, 2026

Wrong Prediction, Right Answer: Recovering Evidence from Collapsed LLM Sequence Scores

When a large language model fails a reasoning task, it is often assumed to lack the underlying capability. However, this conflates a genuine absence of reasoning with a late-stage output bottleneck. We observe a consistent readout gap across diverse reasoning benchmarks: hidden-state probes successfully decode correct answers even when native sequence scoring completely collapses due to structural biases. To test whether instance-specific logic survives this collapse, we introduce a diagnostic protocol using a minimal, target-label-free additive correction. Fitting just two parameters on as few as 25 unlabeled examples recovers 9--34 accuracy points for Qwen3.5 models, transferring successfully to OLMo-2-1B and Llama-3.1-8B. Crucially, these recovered decisions persist on hard instances unresolved by simple lexical overlap and significantly exceed count-preserving permutation baselines. Our results show that many apparent zero-shot reasoning deficits are expression failures masking intact internal logic, urging a narrower interpretation of benchmark evaluations.

Qiyao Yan, Chenpeng Wang, Liangming Pan

State Key Laboratory of Multimedia Information Processing, Peking University · School of Computer Science, Peking University · YiXin-AILab, YIXIN, Beijing, China +1

Language Modeling Language Model Calibration Implicit Reasoning in Language Models LLM Evaluation LLM Reasoning Language Model Probing+1

Aug 30, 2026

Detecting Hidden Chain-of-Thought in Large Language Models with Linguistic, Behavioral, and Mechanistic Indicators

Large language models often answer complex reasoning questions without revealing intermediate steps, raising whether they reason latently or complete patterns. We propose the Hidden CoT Detection Score (HCDS), a comparative behavioral and mechanistic signal measuring whether neutral-prompt behavior aligns more closely with explicit CoT or explicit no- CoT. Here, hidden CoT operationally denotes this neutral-prompt CoT-like alignment; HCDS does not directly observe or prove an unexposed reasoning trace. On GSM8K, HCDS is significantly positive for both Qwen3-4B variants (Thinking

+1.87

,

p = 1.2 \times 10^{-7}

; Instruct

+1.41

,

p = 1.9 \times 10^{-4}

), replicates across a different inference stack and quantization within

0.08

(

+1.80

and

+1.45

), and is not significantly positive in seven of eight length-adjusted calibration-control cells. The unadjusted score produces large positive scores on single-step arithmetic and numeric factual lookup. The variants also respond differently to no-CoT instructions: Instruct complies from the prompt alone, whereas Thinking continues reasoning and requires intervention. These findings show stronger, less prompt-conditional CoT-like behavior in the reasoning-tuned model, consistent with but not proof of latent reasoning. HCDS thus investigates latent reasoning without relying on models' self-reported traces.

Armaan Singh, Ryan Trinh Le, Jasmine Kaur +5

1Stanford University · 2York University · 3Facebook

Implicit Reasoning in Language Models Language Model Probing Large Reasoning Models LLM Interpretability LLM Evaluation Language Model Reasoning Evaluation+1

Aug 18, 2026

ClosureBench: A Constructive Benchmark for Compositional Graph Reasoning

Large language models fail on multi-step compositional reasoning, but measuring that failure is hard, because new models are trained on the benchmarks used to evaluate them. A fixed test set becomes a memorisation check soon after release. Constructive benchmarks avoid this by generating instances on demand. We introduce ClosureBench, a constructive benchmark for graph-relational logical reasoning. Each task is built from explicit primitives (reachability, degree, set operations, connectivity, aggregation), and its reference answer is computed by executing code that implements that logic exactly. Ground truth is therefore verified, and the supply of fresh instances is unlimited. The benchmark spans 26 task categories at three compositional levels, with three independent difficulty axes: graph size, edge density, and query depth. We evaluate models from 1.5B open weights to frontier systems (o3, GPT-4.1, Gemini 2.5, Claude Sonnet 4). Accuracy falls as graph size and query depth increase, and the two axes interact. The difficulty does not lie in the surface form, since it persists when the graph is given as a JSON edge list or an adjacency matrix rather than prose, nor in the reasoning rule, which models state correctly. It lies in carrying that rule out over the graph across many steps. A 4B model fine-tuned to emit verified programs instead of answers stays nearly flat across compositional levels, while every frontier model degrades. o3 falls from 96% on atomic queries to 82% on the most compositional; the 4B model holds at 93% at a fraction of the token cost. The program offloads multi-step execution to a runtime, and the model's remaining errors are almost entirely misread edges. Constructive generation also supports a direct memorisation check, comparing accuracy on seen and fresh instances.

Stefano Goria

AIM Research Lab

LLM Reasoning with Graphs Language Model Reasoning Evaluation Graph Reasoning LLM Reasoning

Aug 18, 2026

Thinking in a Low-Resource Language: What SFT Builds, What RL Fixes, What Accuracy Cannot See

Take three frontier mixture-of-experts models (Alibaba, OpenAI, NVIDIA; 3.6-4.0B active parameters each) and fine-tune them to reason in a low-resource language. On accuracy benchmarks almost nothing happens, and the benchmark itself is noise at this scale: changing only the random seed moves the score by 7.7 points, more than every data and recipe effect we measured. That null is our first result. The real changes live where accuracy cannot see. Base models never think in Greek: 0 of 1,000 reasoning traces, even when the question is Greek, so the model answers correctly while reasoning in a form its user cannot read, audit, or correct. After supervised fine-tuning (SFT), every released checkpoint reasons in the language of the question on ~98% of items, one family at 3x fewer tokens, with judged grammaticality improving on all four models and general ability within a few points of each base: nothing was forgotten, and fluency was gained. We propose six behavioural dimensions that make such changes measurable, each gated to reject any metric that correlates with output length, and we report how our own instruments lied: six failures, each caught by a control. What SFT cannot do is fix its own defects: a quarter of answers skip the requested format, answers leak into the reasoning channel, and an explicit "think in English" is obeyed under half the time. Reinforcement learning with verifiable rewards, pre-registered before training, fixes the first two outright (fallback 24% to 2.5%, leak 3.5% to 0.0%, both against a flat random-reward control) and moves the third (+9.1pp), while the Greek reasoning habit survives an accuracy-only gradient untouched. We release five checkpoints. The instruments, the controls and the pre-registration travel to any low-resource language; Greek is the case that let us measure them.

Ayoub Kirouane, Christos Petrocheilos

Sophea AI, KIEFER SA, Athens, Greece

Fine-Tuning Language Models Structured Reasoning in Language Models Language Model Reasoning Evaluation Supervised Fine-Tuning Low-Resource Language Processing RL for Language Models

Aug 13, 2026

Beyond Local Accuracy: A Protocol-Level Identifiability Audit for Controlled LLM Reasoning Evaluation

LLM benchmark scores can be precise even when the observation protocol does not identify the behavioral property they are intended to measure. In a controlled, solver-grounded setting, we formalize a protocol-level identifiability audit over a finite behavioral policy class: given policies H, observation support O, and estimand

τ

, we test whether O separates every pair with different

τ

. The audit requires zero model calls and resolves our diagnostic case: base-only observation collapses seven frozen deterministic policies into one equivalence class; full support yields seven classes and no cross-estimand collisions; every leave-one-out support retains a constructive collision witness. Empirically, both constrained-generation variants have pair-validity 1.0, yet base accuracy and selective-response fidelity diverge - 0.620 versus 0.324 across six balanced oracle-transition directions (cluster-bootstrap 95% CI [0.600, 0.642] vs. [0.304, 0.345]) - and the gap recurs on a second deterministic source (0.646 vs. 0.331). The audit also synthesizes a minimum identifying support

O^*

for the frozen policy class: two cells instead of the full 36-cell tensor. This case shows how evaluation-design validity can be checked structurally before model inference and why base correctness does not determine intervention-response fidelity.

Junhao Luo, Ning Huang, Ziqi Sha +2

School of Statistics and Data Science, Southwestern University of Finance and Economics

LLM Evaluation LLM Auditing Causal Identifiability Benchmark Validity Language Model Reasoning Evaluation

Aug 12, 2026

Who Thinks Best Depends on How Long You Let Them: Budget-Dependent Rankings in LLM Evaluation

Standard evaluation of large language models assumes stable model rankings across inference conditions. We challenge this assumption by varying the token generation budget, i.e., the maximum tokens a model may produce, across seven levels (64--4,096), evaluating four models on three reasoning benchmarks (56,476 inferences). We report four findings: (i) 3--19% of items exhibit non-monotone behavior (accuracy decreasing with more budget), even after controlling for truncation, and this phenomenon is model-specific (cross-model overlap: 6--14%). (ii) Model rankings reverse across budgets on all benchmarks (

p {<} 0.01

, McNemar). (iii) Oracle analysis reveals model complementarity up to

+27.8

pp, most pronounced at constrained budgets. (iv) A budget-aware router captures 14.1% of the oracle gap cross-domain; budget features help within-domain (

+1.6

to

+5.7

pp) but are domain-specific and hurt transfer (

-1.2

pp). These results argue for budget-conditioned evaluation protocols.

Rodrigo Guedes de Souza, Alison R. Panisson

Federal University of Santa Catarina (UFSC) Brazil

LLM Evaluation Language Model Generation Evaluation Inference-Time Scaling LLM Routing LLM Reasoning Language Model Reasoning Evaluation

Jul 22, 2026

Reference-Free Evaluation of Reasoning in Open-Ended Question Answering

AI-generated answers in high-stakes domains are often fluent but difficult to verify, especially when they contain multi-step reasoning rather than a single final answer. We propose a reasoning-based, reference-free framework for auditing LLM-generated outputs. The method decomposes a generated reasoning trace into segments, labels local premise-target relations using Natural Language Inference (NLI), and organizes these relations into a hypergraph. A deterministic backward AND-OR search then assigns segment-level audit labels that indicate how each segment is grounded within the generated response. We evaluate the framework in two settings: deductive mathematical reasoning with Hard2Verify, and open-ended medical reasoning with UroReason, a new physician-annotated benchmark of LLM reasoning traces from real clinical cases. Across these settings, our NLI-hypergraph audit provides a more reliable reference-free evaluation signal than direct LLM-as-judge baselines. In the clinical setting, state-of-the-art LLM judges often fail to identify problematic reasoning segments, over-accepting fluent but weakly grounded responses. Our results show that QA evaluation should account for how inferential relations compose across a reasoning trace, rather than relying only on final answers or LLMs as verifiers. UroReason will be made available through an API, and our code will be released as open source.

Guneet Singh Kohli, Yuxiang Zhou, Michael Sejr Schlichtkrull +2

Queen Mary University of London · Temple University

LLM-as-a-Judge Reasoning Evaluation LLM Auditing Clinical Language Model Evaluation Language Model Reasoning Evaluation