Daily Papers

Recent advancements in large language models (LLMs) suggest great promises in code and proof generations. However, scaling automated formal verification to real-world projects requires resolving cross-module dependencies and global contexts, which are crucial challenges overlooked by existing LLM-based methods with a special focus on targeting isolated, function-level verification tasks. To systematically explore and address the significant challenges of verifying entire software repositories, we introduce RVBench, the first verification benchmark explicitly designed for repository-level evaluation, constructed from four diverse and complex open-source Verus projects. We further introduce RagVerus, an extensible framework that synergizes retrieval-augmented generation with context-aware prompting to automate proof synthesis for multi-module repositories. RagVerus triples proof pass rates on existing benchmarks under constrained model inference budgets, and achieves a 27% relative improvement on the more challenging RVBench benchmark, demonstrating a scalable and sample-efficient verification solution.

Scaling automated formal verification to real-world projects requires resolving cross-module dependencies and global contexts, which are challenges overlooked by existing function-centric methods. We introduce RagVerus, a framework that synergizes retrieval-augmented generation with context-aware prompting to automate proof synthesis for multi-module repositories, achieving a 27% relative improvement on our novel RepoVBench benchmark -- the first repository-level dataset for Verus with 383 proof completion tasks. RagVerus triples proof pass rates on existing benchmarks under constrained language model budgets, demonstrating a scalable and sample-efficient verification.

Automated formal verification remains challenging for large language models because data for proof assistants and verification-aware languages is scarce, and correctness depends on satisfying precise machine-checkable specifications rather than producing plausible code. This thesis studies how verifier environments can improve LLM generation of verified programs and proofs through reinforcement learning from verifiable rewards (RLVR) and verifier-guided inference-time search. First, we train open-source models in Dafny with RLVR using Group Relative Policy Optimization (GRPO) and related variants, assembling generated candidates into complete programs and scoring them with compiler and verifier outcomes. Initial experiments on an APPS-derived Dafny dataset increased verified reward from 2.2% to 58.1%, but revealed specification hacking, where models exploit weak formal specifications instead of implementing the intended solutions. After filtering underspecified and vulnerable tasks, multi-turn RLVR on the refined benchmark improves the verified pass rate from 9.7% to 31.1%. Second, we develop a verifier-guided inference scaffold in Lean that treats proof generation as structured search over decomposed subgoals, verifier feedback, diagnostics, and repair. With a fixed base model, the full scaffold with proof reviser improves pass rate on an initial VeriCoding pilot set from 46.2% under direct repair to 69.2%. On the larger VERINA dataset, whole-task decomposition plus proof reviser solves 7 of 42 previously unsolved tasks. We also introduce Dalek-Bench, a repository-scale Lean benchmark derived from the Rust curve25519-dalek verification project; preliminary results remain weak, indicating that stronger progress evaluation and task-specific tool-use policies are still needed.

Neural Theorem Proving (NTP) employs LLMs to automate formal proofs in proof assistants. While LLMs have achieved relatively remarkable success in informal reasoning tasks using natural languages, the transition to mechanized formal theorem proving presents persistent challenges. Mechanized proof languages often contain many syntactic constructs and diverse, specialized proof tactics, which facilitate expert use but have no direct counterpart in informal mathematical proofs. These prover-specific idioms represent an additional burden for LLM-based NTPs that might be otherwise successful in generating informal proofs. Seeking to bridge this gap between formal proof construction and informal reasoning, in order to better facilitate NTP, this work approaches these challenges from a language design perspective. We look at common reasoning patterns in informal proofs and in existing mechanized proofs, and design Minilang -- a minimalist proof language that captures these reasoning patterns. In contrast to proof languages (informal and formal) that often feature a large collection of operations with unclear semantic boundaries, Minilang is deliberately kept minimalist -- its core design comprises only 10 operations, each with clear semantic distinctions. We further develop a rule-based translator from Isabelle's language (Isar) to Minilang, translating ~340K existing proofs with an ~85% success rate. Using this translated corpus, we finetune two LLMs to compare machine learning performance on Minilang versus the original Isar. Experiments show Minilang benefits the two LLMs by improving the pass@1 success rate on the PISA benchmark by up to 20/29 percentage points in comparison to the Isar-based LLMs w/wo Sledgehammer. The pass@1 rate reaches 69.1%, exceeding the prior work Baldur's pass@64 (65.7%); the pass@8 rate reaches 79.2%, exceeding the SOTA on PISA (71.0%) achieved by Magnushammer.

Autonomous coding agents increasingly execute multi-step software work, but lifecycle states such as reviewed, tested, DONE, and ready-to-merge remain claims unless supported by current evidence. We present Proof-or-Stop Lifecycle Control, a method that permits lifecycle transitions only when fresh, tracked-source-state-bound, mechanically verifiable evidence satisfies the relevant gate. The method treats agent outputs as claims rather than lifecycle state, and uses proof operationally to mean gate-admissible evidence under a stated trust model, not semantic program correctness. We evaluate an open-source implementation through mechanism tests, a powered control-policy ablation, and operated self-application evidence. The unattended-loop engine passed 10 of 10 scenarios with zero false-DONE, and local-key receipt bundles rejected 18 tamper classes with zero false accepts. In a 9,240-cell ablation, the pre-registered A4 versus A2-prime comparison reduced visible-pass/hidden-fail amplification from 31 of 1,800 injected cells under a compute-budgeted naive loop to 2 of 1,800 under the gated loop, a 1.6 percentage-point improvement in not-amplified rate with a 95 percent confidence interval of [0.8, 2.5]. A near-compute A3 versus A4 comparison, 14 of 1,800 versus 2 of 1,800, indicates that the gain is associated with enforcing review as a lifecycle gate rather than merely adding a reviewer. The self-application corpus contains 565 stories and 1,007 review findings, with 94.8 percent resolved, plus a 68-row high/critical cross-vendor exhibit. These results support Proof-or-Stop as a model-agnostic, host-neutral control layer for deciding which autonomous-agent claims a lifecycle may act on. The evaluation is limited to one model family, 24 ablation tasks, and a self-hosted corpus.

Language models have become increasingly powerful tools for formal mathematical reasoning. However, most existing approaches rely exclusively on either large general-purpose models or smaller specialized models, each with distinct limitations, while training specialized large models still requires significant computational resources. This paper introduces ProofCompass, a novel hybrid methodology that achieves remarkable computational efficiency by strategically guiding existing specialized prover methods, such as DeepSeek-Prover-v1.5-RL (DSP-v1.5) with a Large Language Model (LLM) without requiring additional model training. The LLM provides natural language proof strategies and analyzes failed attempts to select intermediate lemmas, enabling effective problem decomposition. On the miniF2F benchmark, ProofCompass demonstrates substantial resource efficiency: it outperforms DSP-v1.5 (54.9% rightarrow 55.3%) while using 25x fewer attempts (3200 rightarrow 128). Our synergistic approach paves the way for simultaneously improving computational efficiency and accuracy in formal theorem proving.

Recent advances in automated theorem proving (ATP) through LLMs have highlighted the potential of formal reasoning with Lean 4 codes. However, ATP has not yet be revolutionized by the recent posttraining scaling as demonstrated by Open AI O1/O3 and Deepseek R1. In this work, we investigate the entire posttraining of ATP, aiming to align it with breakthroughs in reasoning models in natural languages.To begin, we continual train current ATP models with a hybrid dataset, which consists of numerous statement-proof pairs, and additional data aimed at incorporating cognitive behaviors that emulate human reasoning and hypothesis refinement. Next, we explore reinforcement learning with the use of outcome reward returned by Lean 4 compiler. Through our designed continual training and reinforcement learning processes, we have successfully improved existing formal provers, including both DeepSeek-Prover-v1.5 and Goedel-Prover, achieving state-of-the-art performance in the field of whole-proof generation. For example, we achieve a 59.8% pass rate (pass@32) on MiniF2F. This is an on-going project and we will progressively update our findings, release our data and training details.

Theorem proving is a fundamental task in mathematics. With the advent of large language models (LLMs) and interactive theorem provers (ITPs) like Lean, there has been growing interest in integrating LLMs and ITPs to automate theorem proving. In this approach, the LLM generates proof steps (tactics), and the ITP checks the applicability of the tactics at the current goal. The two systems work together to complete the proof. In this paper, we introduce DS-Prover, a novel dynamic sampling method for theorem proving. This method dynamically determines the number of tactics to apply to expand the current goal, taking into account the remaining time compared to the total allocated time for proving a theorem. This makes the proof search process more efficient by adjusting the balance between exploration and exploitation as time passes. We also augment the training dataset by decomposing simplification and rewrite tactics with multiple premises into tactics with single premises. This gives the model more examples to learn from and helps it to predict the tactics with premises more accurately. We perform our experiments using the Mathlib dataset of the Lean theorem prover and report the performance on two standard datasets, MiniF2F and ProofNet. Our methods achieve significant performance gains on both datasets. We achieved a state-of-the-art performance (Pass@1) of 14.2% on the ProofNet dataset and a performance of 29.8% on MiniF2F, slightly surpassing the best-reported Pass@1 of 29.6% using Lean.

Recent advancements in large language models (LLMs) have sparked considerable interest in automated theorem proving and a prominent line of research integrates stepwise LLM-based provers into tree search. In this paper, we introduce a novel proof-state exploration approach for training data synthesis, designed to produce diverse tactics across a wide range of intermediate proof states, thereby facilitating effective one-shot fine-tuning of LLM as the policy model. We also propose an adaptive beam size strategy, which effectively takes advantage of our data synthesis method and achieves a trade-off between exploration and exploitation during tree search. Evaluations on the MiniF2F and ProofNet benchmarks demonstrate that our method outperforms strong baselines under the stringent Pass@1 metric, attaining an average pass rate of 60.74% on MiniF2F and 21.18% on ProofNet. These results underscore the impact of large-scale synthetic data in advancing automated theorem proving.

Frontier deep research agents (DRAs) plan a research task, synthesize across documents, and return a structured deliverable on demand. They are being deployed in enterprise workflows faster than they are being evaluated. Existing benchmarks measure factual recall, single-hop QA, or generic agentic skill, missing the multi-document, decision-grade work DRAs are deployed to produce. We introduce a benchmark targeting the structured analytical deliverables that fill a management consultant's typical week. We grade three frontier agents, namely Claude Opus 4.6 with web search, OpenAI o3-deep-research, and Google Gemini 3.1 Pro deep-research, on 42 SME-authored prompts. Each of the 126 responses is scored on two layers: deterministic ground-truth verifiers (mean 13.8 per task) and a five-criterion 0-3 SME rubric, composed into a Verifier-Rubric Score (VRS) on 0-100. Most prompts embed cognitive traps that penalize surface-pattern matching. Acceptance under our joint threshold (rubric mean >= 2.5 and verifier rate >= 80%) is uniformly low: Gemini 21.4%, o3 9.5%, Claude 9.5%. Mean VRS scores agree with published rubric-based benchmarks (our top 62.6 vs. APEX-v1 64.2, ProfBench 65.9, ResearchRubrics < 68%), validating the rubric construct. ACCEPT rates sit below APEX-Agents' MC-segment Pass@1 band (12.3-22.7%) on dedicated DR agents; our floor is three points lower despite the harness advantage, opened by stricter conjunctive grading and trap design. Each agent fails distinctively. Claude produces the deliverable most reliably (4.5x the others' rate on file-required tasks) but carries the highest fabrication signature. o3 has the cleanest reasoning average yet drops required sections and propagates arithmetic errors. Gemini is bimodal, with the highest acceptance rate alongside the most zero-scored rubric cells.

Agent systems often decompose a task across multiple roles, but these roles are typically specified by prompts rather than enforced by access controls. Without enforcement, a team pass rate can mask whether agents actually coordinated or whether one role effectively did another role's work. We present TeamBench, a benchmark with 851 task templates and 931 seeded instances for evaluating agent coordination under operating system-enforced role separation. TeamBench separates specification access, workspace editing, and final certification across Planner, Executor, and Verifier roles, so that no role can read the full requirements, modify the workspace, and certify the final answer. Prompt-only and sandbox-enforced teams reach statistically indistinguishable pass rates, but prompt-only runs produce 3.6 times more cases where the verifier attempts to edit the executor's code. Verifiers approve 49% of submissions that fail the deterministic grader, and removing the verifier improves mean partial score in the ablation. Team value is also conditional. Teams benefit when single agents struggle, but hurt when single agents already perform well. A 40-session human study under the same role separation shows that our benchmark exposes interaction patterns that pass rate misses. Solo participants work through the task directly, human participants paired with agents often collapse into quick approval, and human teams spend more effort coordinating missing information across roles.

Large Language Models (LLMs) have recently advanced the field of Automated Theorem Proving (ATP), attaining substantial performance gains through widely adopted test-time scaling strategies, notably reflective Chain-of-Thought (CoT) reasoning and increased sampling passes. However, they both introduce significant computational overhead for inference. Moreover, existing cost analyses typically regulate only the number of sampling passes, while neglecting the substantial disparities in sampling costs introduced by different scaling strategies. In this paper, we systematically compare the efficiency of different test-time scaling strategies for ATP models and demonstrate the inefficiency of the current state-of-the-art (SOTA) open-source approaches. We then investigate approaches to significantly reduce token usage and sample passes while maintaining the original performance. Specifically, we propose two complementary methods that can be integrated into a unified EconRL pipeline for amplified benefits: (1) a dynamic Chain-of-Thought (CoT) switching mechanism designed to mitigate unnecessary token consumption, and (2) Diverse parallel-scaled reinforcement learning (RL) with trainable prefixes to enhance pass rates under constrained sampling passes. Experiments on miniF2F and ProofNet demonstrate that our EconProver achieves comparable performance to baseline methods with only 12% of the computational cost. This work provides actionable insights for deploying lightweight ATP models without sacrificing performance.

We present MaxProof, a population-level test-time scaling framework for competition-level mathematical proof in the MiniMax-M3 series. M3 first trains three proof-oriented capabilities -- proof generation, proof verification, and critique-conditioned proof repair -- using a defense-in-depth generative verifier engineered for low false-positive rate. These capabilities are merged into a single released M3 model. At test time, MaxProof treats the model as a generator, verifier, refiner, and ranker, searches over a population of candidate proofs, and returns one final proof through tournament selection. With MaxProof test-time scaling, the M3 model reaches 35/42 on IMO 2025 and 36/42 on USAMO 2026, exceeding the human gold-medal threshold on both.

Large language models (LLMs) have achieved remarkable performance on high-school and olympiad-style mathematics, yet their capabilities on advanced mathematics remain poorly understood. Existing benchmarks, however, fall short in both scope and evaluation granularity: they provide limited disciplinary coverage and often rely on final-answer correctness or coarse judgments, leaving the validity of the reasoning process inadequately assessed. To bridge this gap, we introduce AdvancedMathBench, a benchmark suite designed to evaluate advanced mathematical reasoning capabilities. Its core proof-generation benchmark, ProverBench, contains 296 problems spanning undergraduate and doctoral qualifying-exam levels. To provide reliable evaluation of the proofs, we develop a dedicated automatic verification pipeline trained on large-scale expert annotations to produce both correctness verdicts and fine-grained assessments of proof errors, which exhibits strong agreement with human experts on held-out proof trajectories. We further introduce VerifierBench, consisting of 888 model-generated proof trajectories paired with expert ground truth, to evaluate whether models can correctly judge proof validity and provide sound verification rationales. Experiments show that AdvancedMathBench remains challenging for frontier models. On proof generation, the best-performing model, GPT-5.5-xhigh, achieves only 75.8 and 66.1 on the UGD and QE splits, respectively, indicating substantial room for improvement on advanced mathematical proof construction. On proof verification, the best model attains a Balanced F1 of only 65.1, and models generally exhibit low true negative rates, suggesting that critical error detection remains a major bottleneck.

Large-scale verifiable prompts underpin the success of Reinforcement Learning with Verifiable Rewards (RLVR), but they contain many uninformative examples and are costly to expand further. Recent studies focus on better exploiting limited training data by prioritizing hard prompts whose rollout pass rate is 0. However, easy prompts with a pass rate of 1 also become increasingly prevalent as training progresses, thereby reducing the effective data size. To mitigate this, we propose Composition-RL, a simple yet useful approach for better utilizing limited verifiable prompts targeting pass-rate-1 prompts. More specifically, Composition-RL automatically composes multiple problems into a new verifiable question and uses these compositional prompts for RL training. Extensive experiments across model sizes from 4B to 30B show that Composition-RL consistently improves reasoning capability over RL trained on the original dataset. Performance can be further boosted with a curriculum variant of Composition-RL that gradually increases compositional depth over training. Additionally, Composition-RL enables more effective cross-domain RL by composing prompts drawn from different domains. Codes, datasets, and models are available at https://github.com/XinXU-USTC/Composition-RL.

Benchmarks like MMLU suggest flagship language models approach factuality saturation above 90\%. LLMpedia shows this picture is incomplete. We materialize {sim}1.3M encyclopedia articles entirely from parametric memory across three model families, then audit every claim against Wikipedia and curated web evidence. For gpt-5-mini, the verifiable true rate is 68.4\% on Wikipedia-covered subjects - more than 21\,pp below MMLU - and the gap is driven by unverifiability (30.5\%), not refutation (1.2\%). Beyond Wikipedia, frontier articles audited against curated web evidence reach 57.6\%; Wikipedia covers only 56.7\% of model-surfaced subjects, and three model families overlap in just 7.3\% of subject choices. In a retrieval-trap benchmark inspired by prior analysis of Grokipedia, LLMpedia is more factual at roughly half the textual similarity to Wikipedia. Every prompt, article, and verdict is released. Data, code, interface: https://llmpedia.net.

Advances in training, post-training, and inference-time methods have enabled frontier reasoning models to win gold medals in math competitions and settle challenging open problems. Gaining trust in the responses of these models requires that natural language proofs be checked for errors. LLM judges are increasingly being adopted to meet the growing demand for evaluating such proofs. While verification is considered easier than generation, what model capability does reliable verification actually require? We systematically evaluate four open-source and two frontier LLMs on datasets of human-graded natural language proofs of competition-level problems. We consider two key metrics: verifier accuracy and self-consistency (the rate of agreement across repeated judgments on the same proof). We observe that smaller open-source models are only up to ~10% behind frontier models in accuracy but they are up to ~25% more inconsistent. Furthermore, we see that verifier accuracy is sensitive to prompt choice across all models. We then demonstrate that the smaller models, in fact, do possess the mathematical capabilities to verify proofs at the level of frontier models, but they struggle to reliably elicit these capabilities with general judging prompts. Through an LLM-guided prompt search, we synthesize an ensemble of specialized prompts that overcome the specific failure modes of smaller models, boosting their performance by up to 9.1% in accuracy and 15.9% in self-consistency. These gains are realized across models and datasets, allowing models like Qwen3.5-35B to perform on par with frontier models such as Gemini 3.1 Pro for proof verification.

We study the process through which reasoning models trained with reinforcement learning on verifiable rewards (RLVR) can learn to solve new problems. We find that RLVR drives performance in two main ways: (1) by compressing pass@k into pass@1 and (2) via "capability gain" in which models learn to solve new problems that they previously could not solve even at high k. We find that while capability gain exists across model scales, learning to solve new problems is primarily driven through self-distillation. We demonstrate these findings across model scales ranging from 0.5B to 72B parameters on >500,000 reasoning problems with prompts and verifiable final answers across math, science, and code domains. We further show that we can significantly improve pass@k rates by leveraging natural language guidance for the model to consider within context while still requiring the model to derive a solution chain from scratch. Based of these insights, we derive Guide -- a new class of online training algorithms. Guide adaptively incorporates hints into the model's context on problems for which all rollouts were initially incorrect and adjusts the importance sampling ratio for the "off-policy" trajectories in order to optimize the policy for contexts in which the hints are no longer present. We describe variants of Guide for GRPO and PPO and empirically show that Guide-GRPO on 7B and 32B parameter models improves generalization over its vanilla counterpart with up to 4% macro-average improvement across math benchmarks. We include careful ablations to analyze Guide's components and theoretically analyze Guide's learning efficiency.

Recent progress in formal theorem proving has benefited from large-scale proof generation and verifier-aware training, but agentic proving is rarely integrated into prover training, appearing only at inference time. We present OProver, a unified framework for agentic formal theorem proving in Lean 4, in which failed proof attempts are iteratively revised using retrieved compiler verified proofs and Lean compiler feedback. OProver is trained through continued pretraining followed by iterative post-training: each iteration runs agentic proving, indexes newly verified proofs into OProofs and the retrieval memory, uses repair trajectories as SFT data, and uses unresolved hard cases for RL. OProofs is built from public Lean resources, large-scale proof synthesis, and agentic proving traces, containing 1.77M Lean statements, 6.86M compiler-verified proofs, and serialized trajectories with retrieved context, failed attempts, feedback, and repairs. Across five benchmarks, OProver-32B attains the best Pass@32 on MiniF2F (93.3%), ProverBench (58.2%), and PutnamBench (11.3%), and ranks second on MathOlympiad (22.8%) and ProofNet (33.2%) more top placements than any prior open-weight whole-proof prover.

When should an AI system's answer be trusted? Formal proof assistants offer certainty but cannot reach most of the problem distribution; scalar LLM judges offer coverage but produce opaque scores that cannot be audited after the fact and are subject to the same coherence issues as any LLM. We present Theoria, a verification architecture that closes this gap. A candidate solution is rewritten into a sequence of typed state transitions, each licensed by an explicit justification, whether that be a citation, computation, or problem-given fact, and every transition is independently auditable. The foundational invariant is completeness of change: every difference between consecutive proof states must be accounted for, so hidden premises surface as unlicensed mutations rather than passing silently. On HLE-Verified Gold (185 text-only expert problems), Theoria certifies 105 at 91.4% strict precision (Wilson 95% CI [84.5%, 95.4%]). Every certification produces a human readable proof trace in which each step can be independently challenged. Holistic LLM judges achieve comparable precision at matched coverage but fail on different problems (Jaccard 0.14-0.36), making the approaches complementary. On 95 adversarial poisoned proofs across 15 domains, structured judges catch 94.7% versus 83.2% for holistic judging (p= 0.0017). The overall 11.5 pp gap concentrates in hidden premises (90.6% vs. 62.5%, a 28 pp difference) and fabricated citations (100% vs. 90%), the error classes where the formal analysis predicts an advantage; performance is identical on arithmetic and theorem-misapplication errors, where no advantage is predicted. On GPQA Diamond (n= 65), certified precision is 97.1% (Wilson CI [85.1%, 99.5%]).

Recent advances in large language models (LLMs) for mathematical reasoning have largely focused on tasks with easily verifiable final answers; however, generating and verifying natural language math proofs remains an open challenge. We identify the absence of a reliable, fine-grained evaluator for LLM-generated math proofs as a critical gap. To address this, we propose a systematic methodology for developing and validating evaluators that assign fine-grained scores on a 0-7 scale to model-generated math proofs. To enable this study, we introduce ProofBench, the first expert-annotated dataset of fine-grained proof ratings, spanning 145 problems from six major math competitions (USAMO, IMO, Putnam, etc) and 435 LLM-generated solutions from Gemini-2.5-pro, o3, and DeepSeek-R1. %with expert gradings. Using ProofBench as a testbed, we systematically explore the evaluator design space across key axes: the backbone model, input context, instructions and evaluation workflow. Our analysis delivers ProofGrader, an evaluator that combines a strong reasoning backbone LM, rich context from reference solutions and marking schemes, and a simple ensembling method; it achieves a low Mean Absolute Error (MAE) of 0.926 against expert scores, significantly outperforming naive baselines. Finally, we demonstrate its practical utility in a best-of-n selection task: at n=16, ProofGrader achieves an average score of 4.14 (out of 7), closing 78% of the gap between a naive binary evaluator (2.48) and the human oracle (4.62), highlighting its potential to advance downstream proof generation.

Large language models have achieved remarkable success on final-answer mathematical problems, largely due to the ease of applying reinforcement learning with verifiable rewards. However, the reasoning underlying these solutions is often flawed. Advancing to rigorous proof-based mathematics requires reliable proof verification capabilities. We begin by analyzing multiple evaluation setups and show that focusing on a single benchmark can lead to brittle or misleading conclusions. To address this, we evaluate both proof-based and final-answer reasoning to obtain a more reliable measure of model performance. We then scale two major generative verification methods (GenSelect and LLM-as-a-Judge) to millions of tokens and identify their combination as the most effective framework for solution verification and selection. We further show that the choice of prompt for LLM-as-a-Judge significantly affects the model's performance, but reinforcement learning can reduce this sensitivity. However, despite improving proof-level metrics, reinforcement learning does not enhance final-answer precision, indicating that current models often reward stylistic or procedural correctness rather than mathematical validity. Our results establish practical guidelines for designing and evaluating scalable proof-verification and selection systems.

We introduce the Formally Verified Automated Programming Progress Standards, or FVAPPS, a benchmark of 4715 samples for writing programs and proving their correctness, the largest formal verification benchmark, including 1083 curated and quality controlled samples. Previously, APPS provided a benchmark and dataset for programming puzzles to be completed in Python and checked against unit tests, of the kind seen in technical assessments in the software engineering industry. Building upon recent approaches for benchmarks in interactive theorem proving, we generalize the unit tests to Lean 4 theorems given without proof (i.e., using Lean's "sorry" keyword). On the 406 theorems of 100 randomly selected samples, Sonnet correctly proves 30% and Gemini correctly proves 18%. We challenge the machine learning and program synthesis communities to solve both each general purpose programming problem and its associated correctness specifications. The benchmark is available at https://huggingface.co/datasets/quinn-dougherty/fvapps.

Large language models for mathematical reasoning are typically trained with outcome-based rewards, which credit only the final answer. In our experiments, we observe that this paradigm is highly susceptible to reward hacking, leading to a substantial overestimation of a model's reasoning ability. This is evidenced by a high incidence of false positives - solutions that reach the correct final answer through an unsound reasoning process. Through a systematic analysis with human verification, we establish a taxonomy of these failure modes, identifying patterns like Miracle Steps - abrupt jumps to a correct output without a valid preceding derivation. Probing experiments suggest a strong association between these Miracle Steps and memorization, where the model appears to recall the answer directly rather than deriving it. To mitigate this systemic issue, we introduce the Rubric Reward Model (RRM), a process-oriented reward function that evaluates the entire reasoning trajectory against problem-specific rubrics. The generative RRM provides fine-grained, calibrated rewards (0-1) that explicitly penalize logical flaws and encourage rigorous deduction. When integrated into a reinforcement learning pipeline, RRM-based training consistently outperforms outcome-only supervision across four math benchmarks. Notably, it boosts Verified Pass@1024 on AIME2024 from 26.7% to 62.6% and reduces the incidence of Miracle Steps by 71%. Our work demonstrates that rewarding the solution process is crucial for building models that are not only more accurate but also more reliable.

We introduce ProofGrid, a benchmark suite for evaluating LLM reasoning through machine-checkable proofs rather than final answers alone. ProofGrid contains 15 tasks spanning proof writing, proof checking, proof masking, and proof gap-filling. Tasks are expressed in minimal formal notation, especially NDL, a compact natural-deduction language that fits in short prompts and supports precise, auditable verification. This yields mechanical, reproducible, and fine-grained evaluation rather than judgments by humans or LLMs. ProofGrid covers a calibrated difficulty spectrum, from foundational reasoning tests to structurally rich challenge tasks that no current model solves, while minimizing reliance on domain knowledge, solver delegation, and long-context artifacts. We also develop a comparative framework for reasoning benchmarks and use it to situate ProofGrid relative to existing work in terms of representation, verification guarantees, and reasoning depth. Methodologically, we introduce an instrumented proof-checking pipeline that tolerates minor surface deviations while locating the first substantive reasoning failure, improving measurement resolution and separating proof planning from low-level execution noise. Using this pipeline, we evaluate a broad range of open and proprietary models. Results show rapid progress but substantial remaining limits: frontier models perform well on several foundational tasks, yet difficult tasks, especially those requiring global combinatorial reasoning or low-level proof synthesis, remain far from solved. We also identify epistemic instability, where models generate flawed proofs yet correctly reject those local inferences in isolation, and formalize this with an Epistemic Stability Index. Finally, we complement accuracy with 2PL IRT analyses, Wright maps, and a normalized task-discrimination measure based on Fisher information.

The high cost of agentic workflows in formal mathematics hinders large-scale data synthesis, exacerbating the scarcity of open-source corpora. To address this, we introduce TheoremForge, a cost-effective formal data synthesis pipeline that decomposes the formalization process into five sub-tasks, which are statement formalization, proof generation, premise selection, proof correction and proof sketching. By implementing a Decoupled Extraction Strategy, the workflow recovers valid training signals from globally failed trajectories, effectively utilizing wasted computation. Experiments on a 2,000-problem benchmark demonstrate that TheoremForge achieves a Verified Rate of 12.6\%, surpassing the 8.6\% baseline, at an average cost of only \0.481 per successful trajectory using Gemini-3-Flash. Crucially, our strategy increases data yield by 1.6\times$ for proof generation compared to standard filtering. These results establish TheoremForge as a scalable framework for constructing a data flywheel to train future expert models. Our code is available https://github.com/timechess/TheoremForge{here}.

Premise selection is a fundamental problem of automated theorem proving. Previous works often use intricate symbolic methods, rely on domain knowledge, and require significant engineering effort to solve this task. In this work, we show that Magnushammer, a neural transformer-based approach, can outperform traditional symbolic systems by a large margin. Tested on the PISA benchmark, Magnushammer achieves 59.5% proof rate compared to a 38.3% proof rate of Sledgehammer, the most mature and popular symbolic-based solver. Furthermore, by combining Magnushammer with a neural formal prover based on a language model, we significantly improve the previous state-of-the-art proof rate from 57.0% to 71.0%.

Large language models (LLMs) can generate plausible code but offer limited guarantees of correctness. Formally verifying that implementations satisfy specifications requires constructing machine-checkable proofs, a task that remains beyond current automation. We propose a hierarchical proof search framework for automated code verification in Lean~4 that decomposes complex verification goals into structurally simpler subgoals before attempting tactic-level proving. Central to our approach is a principled decomposition score that combines constructive justification with structural effectiveness. Crucially, this score serves as both the training reward and the inference-time ranking criterion, ensuring strict alignment between optimization and deployment. We train Goedel-Code-Prover-8B, a single unified policy for both decomposition and completion, via supervised initialization followed by hybrid reinforcement learning, where a continuous decomposition reward drives planning exploration while supervised replay stabilizes proof generation. On three Lean-based code verification benchmarks comprising 427 tasks, our 8B-parameter model achieves a 62.0\% prove success rate, a 2.6times improvement over the strongest baseline, surpassing neural provers up to 84times larger. We further observe consistent inference-time scaling: success rates improve monotonically with search iterations and sampling budget, with our trained model achieving greater efficiency than frontier off-the-shelf models of comparable scale.

Large language models (LLMs) have made remarkable progress in code generation, but competitive programming remains a challenge. Recent training-based methods have improved code generation by using reinforcement learning (RL) with execution feedback. The more recent framework CURE further incorporates test generation into the training process, jointly training a Coder and a Tester within a single model. At inference time, the Coder generates many candidate programs, and the Tester generates tests from the problem description. The candidate who passes the most of the generated tests is selected as the final answer. However, CURE has two critical limitations. First, the Tester never reads any candidate code, so its tests often fail to expose implementation-specific bugs. Second, the Coder generates every candidate from scratch and never learns to fix a buggy program based on a failed test. To address these limitations, we propose FixAudit, which approaches competitive code generation from a new perspective: starting from a single initial candidate, it iteratively improves the candidate through a targeted test-and-repair debugging cycle. The framework trains one shared model with two specialized roles through four stages: the Fixer, which repairs the current candidate based on a failing test, and the Auditor, which reads the candidate code to generate new tests that expose its remaining bugs. We evaluate FixAudit on three benchmarks: APPS, CodeContests, and xCodeEval. Applied to a 7B model, the framework surpasses the average performance of the larger 32B baseline within the same model family under the zero-shot setting. Compared to strong baselines built on the same 7B base model, FixAudit improves average Pass@1 by 35.1% to 36.8% and average AvgPassRatio by 7.1% to 24.5%.

The rise of reasoning models necessitates large-scale verifiable data, for which programming tasks serve as an ideal source. However, while competitive programming platforms provide abundant problems and solutions, high-quality test cases for verification remain scarce. Existing approaches attempt to synthesize test cases using Large Language Models (LLMs), but rely solely on the model's intrinsic generation capabilities without external feedback, frequently resulting in insufficiently diverse cases. To address this limitation, we propose a Feedback-Driven Iterative Framework for comprehensive test case construction. Specifically, our method leverages the LLM to generate initial test cases, executes them against known correct and incorrect solutions, and utilizes the failed results as feedback to guide the LLM in refining the test cases toward high fidelity and discriminability. We then apply this method to the CodeContests dataset to construct an optimized high-quality derivative, CodeContests-O. Evaluating against the entire pool of solutions (1.1 times 10^7 in total), our dataset achieves an average True Positive Rate (TPR) of 89.37% and True Negative Rate (TNR) of 90.89%, significantly outperforming the CodeContests and CodeContests+ by margins of 4.32% and 9.37%, respectively. Furthermore, fine-tuning the Qwen2.5-7B model on CodeContests-O results in a 9.52% improvement on LiveCodeBench (Pass@1). Experiments demonstrate the effectiveness of our framework and the quality of CodeContests-O. To support reproducibility and facilitate future research, we release the https://github.com/cai-jianfeng/CodeContests-O{code} and https://huggingface.co/datasets/caijanfeng/CodeContests-O{dataset}.

Skills are becoming the capability layer through which LLM agents turn plans into actions, but their use introduces security risks such as data leakage, unauthorized operations, and tool misuse. Existing vetting usually evaluates each skill in isolation, while real agent tasks often invoke multiple skills in a shared execution context. This creates Skill Composition Risk (SCR): a skill that appears benign alone can become harmful when its outputs, trust signals, authorization cues, or side effects influence later invocations along an activated path. We introduce SCR-Bench to evaluate this risk in controlled, sandboxed skill environments. Rather than relying only on textual intent or surface behavior, SCR-Bench records downstream state changes and path-level outcomes across composed skill executions. It contains three sub-benchmarks: SCR-CapFlow for capability-flow composition, SCR-TrustLift for trust-transfer composition, and SCR-AuthBlur for authorization-confusion composition. Across SCR-Bench, composed paths expose risks that are largely absent under isolated evaluation. In SCR-CapFlow, attack success rate reaches 33.6 percent under composition, compared with near-zero isolated baselines. In SCR-TrustLift, attack success rate exceeds 96.5 percent on four of five backends. In SCR-AuthBlur, the risky-approval rate increases by 71.8 percent relative to the L0 isolated baseline under the L1 context setting. These results show that agent skill security should be assessed at the level of activated paths rather than isolated artifacts. SCR and SCR-Bench provide a foundation for path-aware risk evaluation and defense in LLM agent skill ecosystems. Benchmark: https://github.com/saint-viperx/SCR_Bench.

We introduce Goedel-Prover, an open-source large language model (LLM) that achieves the state-of-the-art (SOTA) performance in automated formal proof generation for mathematical problems. The key challenge in this field is the scarcity of formalized math statements and proofs, which we tackle in the following ways. We train statement formalizers to translate the natural language math problems from Numina into formal language (Lean 4), creating a dataset of 1.64 million formal statements. LLMs are used to check that the formal statements accurately preserve the content of the original natural language problems. We then iteratively build a large dataset of formal proofs by training a series of provers. Each prover succeeds in proving many statements that the previous ones could not, and these new proofs are added to the training set for the next prover. The final prover outperforms all existing open-source models in whole-proof generation. On the miniF2F benchmark, it achieves a 57.6% success rate (Pass@32), exceeding the previous best open-source model by 7.6%. On PutnamBench, Goedel-Prover successfully solves 7 problems (Pass@512), ranking first on the leaderboard. Furthermore, it generates 29.7K formal proofs for Lean Workbook problems, nearly doubling the 15.7K produced by earlier works.

Automatic LLM benchmarks, such as AlpacaEval 2.0, Arena-Hard-Auto, and MT-Bench, have become popular for evaluating language models due to their cost-effectiveness and scalability compared to human evaluation. Achieving high win rates on these benchmarks can significantly boost the promotional impact of newly released language models. This promotional benefit may motivate tricks, such as manipulating model output length or style to game win rates, even though several mechanisms have been developed to control length and disentangle style to reduce gameability. Nonetheless, we show that even a "null model" that always outputs a constant response (irrelevant to input instructions) can cheat automatic benchmarks and achieve top-ranked win rates: an 86.5% LC win rate on AlpacaEval 2.0; an 83.0 score on Arena-Hard-Auto; and a 9.55 score on MT-Bench. Moreover, the crafted cheating outputs are transferable because we assume that the instructions of these benchmarks (e.g., 805 samples of AlpacaEval 2.0) are private and cannot be accessed. While our experiments are primarily proof-of-concept, an adversary could use LLMs to generate more imperceptible cheating responses, unethically benefiting from high win rates and promotional impact. Our findings call for the development of anti-cheating mechanisms for reliable automatic benchmarks. The code is available at https://github.com/sail-sg/Cheating-LLM-Benchmarks.

As post-training techniques evolve, large language models (LLMs) are increasingly augmented with structured multi-step reasoning abilities, often optimized through reinforcement learning. These reasoning-enhanced models outperform standard LLMs on complex tasks and now underpin many commercial LLM APIs. However, to protect proprietary behavior and reduce verbosity, providers typically conceal the reasoning traces while returning only the final answer. This opacity introduces a critical transparency gap: users are billed for invisible reasoning tokens, which often account for the majority of the cost, yet have no means to verify their authenticity. This opens the door to token count inflation, where providers may overreport token usage or inject synthetic, low-effort tokens to inflate charges. To address this issue, we propose CoIn, a verification framework that audits both the quantity and semantic validity of hidden tokens. CoIn constructs a verifiable hash tree from token embedding fingerprints to check token counts, and uses embedding-based relevance matching to detect fabricated reasoning content. Experiments demonstrate that CoIn, when deployed as a trusted third-party auditor, can effectively detect token count inflation with a success rate reaching up to 94.7%, showing the strong ability to restore billing transparency in opaque LLM services. The dataset and code are available at https://github.com/CASE-Lab-UMD/LLM-Auditing-CoIn.

The rapid advancement of reasoning capabilities in large language models (LLMs) has led to notable improvements on mathematical benchmarks. However, many of the most commonly used evaluation datasets (e.g., AIME 2024) are widely available online, making it difficult to disentangle genuine reasoning from potential memorization. Furthermore, these benchmarks do not evaluate proof-writing capabilities, which are crucial for many mathematical tasks. To address this, we introduce MathArena, a new benchmark based on the following key insight: recurring math competitions provide a stream of high-quality, challenging problems that can be used for real-time evaluation of LLMs. By evaluating models as soon as new problems are released, we effectively eliminate the risk of contamination. Using this framework, we find strong signs of contamination in AIME 2024. Nonetheless, evaluations on harder competitions, such as CMIMC 2025, demonstrate impressive reasoning capabilities in top-performing models. MathArena is also the first benchmark for proof-writing capabilities. On IMO 2025, top models achieve slightly less than 40%, demonstrating both notable progress and significant room for improvement. So far, we have evaluated over 50 models across seven competitions, totaling 162 problems. As an evolving benchmark, MathArena will continue to track the progress of LLMs on newly released competitions, ensuring rigorous and up-to-date evaluation of mathematical reasoning.

Large language models (LLMs) have shown increasing competence in solving mathematical reasoning problems. However, many open-source LLMs still struggle with errors in calculation and semantic understanding during intermediate reasoning steps. In this work, we introduce Prove, a simple yet effective framework that leverages translated programs derived from natural language solutions as a verification mechanism to filter out potentially incorrect reasoning paths before aggregating final answers. Unlike vanilla majority voting, our approach filters out solutions whose corresponding program output is inconsistent with the generated solution, aggregating only those that pass verification. We conducted extensive experiments using 13 open-source LLMs from various model families and sizes, ranging from 0.5B to 13B parameters, across eight mathematical benchmarks. Our results show that Prove consistently outperforms vanilla majority voting as a heuristic for solving mathematical reasoning tasks across all model sizes and datasets, achieving improvements of up to 18% on GSM8K and 8% on MATH-500. Our codes are available at https://github.com/declare-lab/prove.

Reinforcement Learning with Verifiable Rewards (RLVR) trains policies against automated verifiers to avoid costly human labeling. To reduce vulnerability to verifier hacking, many RLVR systems collapse rewards to binary {0,1} during training. This choice carries a cost: it introduces false negatives (rejecting correct answers, FNs) and false positives (accepting incorrect ones, FPs). For instance, a rule-based checker may mark the correct fraction 12{36} as wrong when compared against the canonical 1{3} due to brittle parsing/equivalence rules (FN), while a large language model (LLM) judges can be gamed by superficial cues or even a single adversarial token, yielding inflated correctness for wrong solutions (FP). We formalize verifier unreliability by modeling the verifier as a stochastic reward channel with asymmetric noise rates. From this abstraction, we derive two correction algorithms for verifier errors. The first is a backward correction that de-biases the observed binary reward to recover an unbiased estimator of the clean policy gradient. The second is a forward correction that reweights score-function terms so that the expected update direction aligns with the clean gradient; notably, it requires only the FN rate. We implement both as lightweight hooks in a group relative policy optimization (GRPO)-based RLVR pipeline and evaluate them on math-reasoning models and benchmarks. Across models and datasets, both corrections improve over uncorrected training; the forward variant converges faster and remains stable under heavier noise. Finally, we show a practical appeal mechanism in which a lightweight LLM verifier estimates the FN rate online by rechecking rule-based negatives, obtaining outperformance compared with other state-of-the-art contenders.

Although large language models (LLMs) have recently achieved remarkable performance on various complex reasoning benchmarks, the academic community still lacks an in-depth understanding of base model training processes and data quality. To address this, we construct a large-scale, difficulty-graded reasoning dataset containing approximately 3.34 million unique queries of varying difficulty levels and about 40 million distilled responses generated by multiple models over several passes. Leveraging pass rate and Coefficient of Variation (CV), we precisely select the most valuable training data to enhance reasoning capability. Notably, we observe a training pattern shift, indicating that reasoning-focused training based on base models requires higher learning rates for effective training. Using this carefully selected data, we significantly improve the reasoning capabilities of the base model, achieving a pass rate of 79.2\% on the AIME2024 mathematical reasoning benchmark. This result surpasses most current distilled models and closely approaches state-of-the-art performance. We provide detailed descriptions of our data processing, difficulty assessment, and training methodology, and have publicly released all datasets and methods to promote rapid progress in open-source long-reasoning LLMs. The dataset is available at: https://huggingface.co/datasets/a-m-team/AM-DeepSeek-Distilled-40M

The formalization of existing mathematical proofs is a notoriously difficult process. Despite decades of research on automation and proof assistants, writing formal proofs remains arduous and only accessible to a few experts. While previous studies to automate formalization focused on powerful search algorithms, no attempts were made to take advantage of available informal proofs. In this work, we introduce Draft, Sketch, and Prove (DSP), a method that maps informal proofs to formal proof sketches, and uses the sketches to guide an automated prover by directing its search to easier sub-problems. We investigate two relevant setups where informal proofs are either written by humans or generated by a language model. Our experiments and ablation studies show that large language models are able to produce well-structured formal sketches that follow the same reasoning steps as the informal proofs. Guiding an automated prover with these sketches enhances its performance from 20.9% to 39.3% on a collection of mathematical competition problems.

The widespread adoption of deep neural networks (DNNs) requires efficient techniques for verifying their safety. DNN verifiers are complex tools, which might contain bugs that could compromise their soundness and undermine the reliability of the verification process. This concern can be mitigated using proofs: artifacts that are checkable by an external and reliable proof checker, and which attest to the correctness of the verification process. However, such proofs tend to be extremely large, limiting their use in many scenarios. In this work, we address this problem by minimizing proofs of unsatisfiability produced by DNN verifiers. We present algorithms that remove facts which were learned during the verification process, but which are unnecessary for the proof itself. Conceptually, our method analyzes the dependencies among facts used to deduce UNSAT, and removes facts that did not contribute. We then further minimize the proof by eliminating remaining unnecessary dependencies, using two alternative procedures. We implemented our algorithms on top of a proof producing DNN verifier, and evaluated them across several benchmarks. Our results show that our best-performing algorithm reduces proof size by 37%-82% and proof checking time by 30%-88%, while introducing a runtime overhead of 7%-20% to the verification process itself.

Multi-server MCP agents create an information-flow control problem: faithful tool composition can turn individually benign read/write permissions into cross-boundary credential propagation -- a structural side effect of workflow topology, not necessarily malicious model behavior. We present MCPHunt, to our knowledge the first controlled benchmark that isolates non-adversarial, verbatim credential propagation across multi-server MCP trust boundaries, with three methodological contributions: (1) canary-based taint tracking that reduces propagation detection to objective string matching; (2) an environment-controlled coverage design with risky, benign, and hard-negative conditions that validates pipeline soundness and controls for credential-format confounds; (3) CRS stratification that disentangles task-mandated propagation (faithful execution of verbatim-transfer instructions) from policy-violating propagation (credentials included despite the option to redact). Across 3,615 main-benchmark traces from 5 models spanning 147 tasks and 9 mechanism families, policy-violating propagation rates reach 11.5--41.3% across all models. This propagation is pathway-specific (25x cross-mechanism range) and concentrated in browser-mediated data flows; hard-negative controls provide evidence that production-format credentials are not necessary -- prompt-directed cross-boundary data flow is sufficient. A prompt-mitigation study across 3 models reduces policy-violating propagation by up to 97% while preserving 80.5% utility, but effectiveness varies with instruction-following capability -- suggesting that prompt-level defenses alone may not suffice. Code, traces, and labeling pipeline are released under MIT and CC BY 4.0.

We introduce DeepSeek-Prover-V1.5, an open-source language model designed for theorem proving in Lean 4, which enhances DeepSeek-Prover-V1 by optimizing both training and inference processes. Pre-trained on DeepSeekMath-Base with specialization in formal mathematical languages, the model undergoes supervised fine-tuning using an enhanced formal theorem proving dataset derived from DeepSeek-Prover-V1. Further refinement is achieved through reinforcement learning from proof assistant feedback (RLPAF). Beyond the single-pass whole-proof generation approach of DeepSeek-Prover-V1, we propose RMaxTS, a variant of Monte-Carlo tree search that employs an intrinsic-reward-driven exploration strategy to generate diverse proof paths. DeepSeek-Prover-V1.5 demonstrates significant improvements over DeepSeek-Prover-V1, achieving new state-of-the-art results on the test set of the high school level miniF2F benchmark (63.5%) and the undergraduate level ProofNet benchmark (25.3%).

Test-time reasoning is increasingly used as a serving-time control knob, but extra reasoning is not uniformly valuable: it can repair failed attempts, waste compute on already-correct answers, or introduce harmful answer changes. We study this as a deployment allocation problem rather than a new-verifier problem. We introduce \sevra, Selective Verification for Reasoning Allocation, a serving-layer controller that decides whether to preserve a frozen solver's initial answer or invoke active verification. Using a frozen Qwen3-4B solver, we log intervention outcomes and train recoverability-aware gates from serving-visible attempt state. On \mathfive, selective verification reaches 76.3\% accuracy, compared with 75.5\% for always verifying, while reducing post-generation tokens by 26.8\% and harmful flips from 2.2\% to 1.0\%. However, an 8,192-token initial solve reaches 76.0\% accuracy with 28\% fewer total model tokens, showing that selective recovery is useful but not the best tested cost frontier. In frozen transfer to \gsm, the selective policy verifies only 3.0\% of examples, improves accuracy from 93.4\% to 94.5\%, and reduces verification tokens by 91.2\% relative to always verifying; again, a longer initial solve matches its accuracy with fewer realized tokens. On CommonsenseQA, always-on verification hurts, while Self-Consistency@5 improves accuracy at about five times the realized token cost. The resulting deployment rule is: tune the initial budget first, then use selective recovery when explicit checks, bounded retries, auditability, or regression-risk control matter.

Large language models have achieved strong performance on medical reasoning benchmarks, yet their deployment in clinical settings demands rigorous verification to ensure factual accuracy. While reward models offer a scalable approach for reasoning trace verification, existing methods face two limitations: they produce only scalar reward values without explicit justification, and they rely on single-pass retrieval that precludes adaptive knowledge access as verification unfolds. We introduce method, an agentic framework that addresses these limitations by training medical reasoning verifiers to iteratively query external medical corpora during evaluation. Our approach combines tool-augmented verification with an iterative reinforcement learning paradigm that requires only trace-level supervision, alongside an adaptive curriculum mechanism that dynamically adjusts training data distribution. Across four medical reasoning benchmarks, method achieves substantial gains over existing methods, improving MedQA accuracy by 23.5% and MedXpertQA by 32.0% relative to the base generator in particular. Crucially, method demonstrates an 8times reduction in sampling budget requirement compared to prior reward model baselines. These findings establish that grounding verification in dynamically retrieved evidence offers a principled path toward more reliable medical reasoning systems.

A general-purpose language model that answers a harmful question returns text; a coding model that complies with a malicious request can return a working weapon -- a keylogger, a ransomware stub, an exploit that runs as written. This asymmetry in the severity of a single act of compliance implies coding-specialized models should clear a higher refusal bar than general-purpose chat models, not a lower one, yet the field cannot presently tell whether they do. Refusal benchmarks for malicious code are fragmented: they mix requests for executable software (ready-to-run weapons) with requests for harmful security knowledge (information a human must still operationalise) and report refusal rates over non-comparable corpora, so no single statistic measures the property that actually matters. This paper introduces an expanded consensus-labeled prompt bank that distinguishes between these two request types and provides a construct-stable substrate for cross-corpus coding-model compliance measurement. Eight corpora (ASTRA, CySecBench, AdvBench/harmful_behaviors, JailbreakBench, MalwareBench, RedCode, RMCBench, Scam2Prompt) are consolidated and classified under a five-judge consensus protocol (6,675 prompts x 5 judges = 33,375 calls). The panel reaches Fleiss' kappa = 0.767 [95% CI 0.755, 0.777] ("substantial"); 95.0% of prompts draw at least four agreeing judges, 76.9% are unanimous, and the panel reproduces the earlier four-corpus release at Cohen's kappa = 0.952 on the 3,133 shared prompts. The released bank comprises 4,748 consensus-CODE prompts (executable malicious code requests) and 1,923 consensus-KNOWLEDGE prompts (harmful security knowledge requests). The bank is the validated instrument the field has lacked: a reliability-quantified basis for testing whether coding models meet the stricter refusal standard their executable output demands.

Nearly all jurisdictions in the United States require a professional license exam, commonly referred to as "the Bar Exam," as a precondition for law practice. To even sit for the exam, most jurisdictions require that an applicant completes at least seven years of post-secondary education, including three years at an accredited law school. In addition, most test-takers also undergo weeks to months of further, exam-specific preparation. Despite this significant investment of time and capital, approximately one in five test-takers still score under the rate required to pass the exam on their first try. In the face of a complex task that requires such depth of knowledge, what, then, should we expect of the state of the art in "AI?" In this research, we document our experimental evaluation of the performance of OpenAI's `text-davinci-003` model, often-referred to as GPT-3.5, on the multistate multiple choice (MBE) section of the exam. While we find no benefit in fine-tuning over GPT-3.5's zero-shot performance at the scale of our training data, we do find that hyperparameter optimization and prompt engineering positively impacted GPT-3.5's zero-shot performance. For best prompt and parameters, GPT-3.5 achieves a headline correct rate of 50.3% on a complete NCBE MBE practice exam, significantly in excess of the 25% baseline guessing rate, and performs at a passing rate for both Evidence and Torts. GPT-3.5's ranking of responses is also highly-correlated with correctness; its top two and top three choices are correct 71% and 88% of the time, respectively, indicating very strong non-entailment performance. While our ability to interpret these results is limited by nascent scientific understanding of LLMs and the proprietary nature of GPT, we believe that these results strongly suggest that an LLM will pass the MBE component of the Bar Exam in the near future.

Proof autoformalization, the task of translating natural language theorems and proofs into machine-verifiable code, is a critical step for integrating large language models into rigorous mathematical workflows. Current approaches focus on producing executable code, but they frequently fail to preserve the semantic meaning and logical structure of the original human-written argument. To address this, we introduce ProofFlow, a novel pipeline that treats structural fidelity as a primary objective. ProofFlow first constructs a directed acyclic graph (DAG) to map the logical dependencies between proof steps. Then, it employs a novel lemma-based approach to systematically formalize each step as an intermediate lemma, preserving the logical structure of the original argument. To facilitate evaluation, we present a new benchmark of 184 undergraduate-level problems, manually annotated with step-by-step solutions and logical dependency graphs, and introduce ProofScore, a new composite metric to evaluate syntactic correctness, semantic faithfulness, and structural fidelity. Experimental results show our pipeline sets a new state-of-the-art for autoformalization, achieving a ProofScore of 0.545, substantially exceeding baselines like full-proof formalization (0.123), which processes the entire proof at once, and step-proof formalization (0.072), which handles each step independently. Our pipeline, benchmark, and score metric are open-sourced to encourage further progress at https://github.com/Huawei-AI4Math/ProofFlow.

Large language models (LLMs) can act as both problem solvers and solution verifiers, where the latter select high-quality answers from a pool of solver-generated candidates. This raises the question of under what conditions verification pays off in solver-verifier systems. Prior work has conducted only limited studies of the factors influencing verification performance, focusing primarily on self-verification and examining neither the relationship between solver and verifier model families nor the effects of reasoning post-training. To rectify this, we present a systematic study across 37 models spanning multiple families, sizes, and base vs. post-trained variants, evaluated on 9 benchmarks covering logical reasoning, structured puzzles, symbolic computation, mathematics, commonsense, factual recall, and domain knowledge. In order to support our analysis, we introduce and empirically validate verifier gain, a metric that predicts the performance improvements from test-time verifier-based rejection sampling. Our experiments find that 1) verification across model families is more effective than either self-verification or verification within the same family, and more generally that the benefits of verification decrease as the solver and verifier become more similar, 2) reasoning post-training weakens self-improvement abilities but strengthens cross-family improvement, and 3) some tasks are inherently more amenable to improvement through verification, particularly mathematical and logical tasks.

Reinforcement learning with verifiable rewards (RLVR) has advanced the reasoning capabilities of large language models (LLMs). However, prevailing RLVR methods exhibit a systematic bias toward exploitation over exploration, as evidenced by improved pass@1 but reduced pass@K (K>1) performance. To understand this issue, we analyze training dynamics of RLVR methods by tracking the token-level probability distributions over vocabulary candidates. Our analysis reveals a consistent probability concentration effect where the top-1 candidate increasingly accumulates probability mass and suppresses that of other candidates. More importantly, stronger over-concentration correlates with worse pass@K performance. Inspired by this finding, we propose Simple Pass@K Optimization (SimKO), a method designed to mitigate the over-concentration issue, thereby encouraging exploration. SimKO operates in an asymmetrical manner. For verified-correct responses, it boosts the probabilities of the top-K candidates. For verified-incorrect responses, it applies stronger penalties to the top-1 candidate. We observe that this asymmetric design is particularly effective at mitigating over-concentration when applied at tokens with high entropy. Across various math and logical-reasoning benchmarks, SimKO consistently yields higher pass@K for a wide range of K, providing a simple way to improve RLVR's exploration.

Prompt optimization methods either analyze individual failures in isolation or compare prompt variants across examples, operating on single execution traces with no access to the reasoning process distinguishing success from failure on the same input. We introduce ContraPrompt, built on the observation that when a model fails but succeeds on a retry with feedback, the difference between its two chain-of-thought traces constitutes an optimization signal not captured by prior methods. Unlike prior contrastive methods, we compare complete intermediate reasoning processes: the two traces share model, input, and base prompt, so remaining differences reflect reasoning strategy and appended error feedback -- we call this dyadic reasoning trace analysis. The multi-attempt solving phase is an instrumented agentic retry loop that generates contrastive data automatically without human annotation. Extracted rules are organized into an input-aware decision tree routing instructions by observable input characteristics. On four reasoning and compliance benchmarks, ContraPrompt outperforms GEPA (Agrawal et al., 2026) on all four, with absolute gains of +8.29 pp on HotPotQA (+20.8% rel.), +2.21 pp on GDPR-Bench (+18.2% rel.), +7.14 pp on GPQA Diamond (+10.6% rel.), and +0.74 pp on BBH (+0.85% rel.). Ablations confirm dyadic trace contrastivity is the critical component, with a -16% relative average drop upon its removal. On 53 EvalSet black-box optimization problems, ContraPrompt beats GEPA on 11, ties on 41, and loses on 1 at equal budget. On FiNER-139 financial named entity recognition (Loukas et al., 2022), ContraPrompt achieves +7.77 pp over the unoptimized baseline (+11.6% rel.) and +1.94 pp over GEPA (+2.66% rel.), with branch conditions aligning with standard US GAAP financial-instrument categories.

State-of-the-art (SOTA) LLMs have progressed from struggling on proof-based Olympiad problems to solving most of the IMO 2025 problems, with leading systems reportedly handling 5 of 6 problems. Given this progress, we assess how well these models can grade proofs: detecting errors, judging their severity, and assigning fair scores beyond binary correctness. We study proof-analysis capabilities using a corpus of 90 Gemini 2.5 Pro-generated solutions that we grade on a 1-4 scale with detailed error annotations, and on MathArena solution sets for IMO/USAMO 2025 scored on a 0-7 scale. Our analysis shows that models can reliably flag incorrect (including subtly incorrect) solutions but exhibit calibration gaps in how partial credit is assigned. To address this, we introduce agentic workflows that extract and analyze reference solutions and automatically derive problem-specific rubrics for a multi-step grading process. We instantiate and compare different design choices for the grading workflows, and evaluate their trade-offs. Across our annotated corpus and MathArena, our proposed workflows achieve higher agreement with human grades and more consistent handling of partial credit across metrics. We release all code, data, and prompts/logs to facilitate future research.

Large Language Models (LLMs) often exhibit significant behavioral shifts when they perceive a change from a real-world deployment context to a controlled evaluation setting, a phenomenon known as "evaluation awareness." This discrepancy poses a critical challenge for AI alignment, as benchmark performance may not accurately reflect a model's true safety and honesty. In this work, we systematically quantify these behavioral changes by manipulating the perceived context of prompts. We introduce a methodology that uses a linear probe to score prompts on a continuous scale from "test-like" to "deploy-like" and leverage an LLM rewriting strategy to shift these prompts towards a more natural, deployment-style context while preserving the original task. Using this method, we achieved a 30% increase in the average probe score across a strategic role-playing dataset after rewriting. Evaluating a suite of state-of-the-art models on these original and rewritten prompts, we find that rewritten "deploy-like" prompts induce a significant and consistent shift in behavior. Across all models, we observed an average increase in honest responses of 5.26% and a corresponding average decrease in deceptive responses of 12.40%. Furthermore, refusal rates increased by an average of 6.38%, indicating heightened safety compliance. Our findings demonstrate that evaluation awareness is a quantifiable and manipulable factor that directly influences LLM behavior, revealing that models are more prone to unsafe or deceptive outputs in perceived test environments. This underscores the urgent need for more realistic evaluation frameworks to accurately gauge true model alignment before deployment.

Solving math problems through verifiable languages such as Lean has significantly impacted both the mathematics and computer science communities. Current state-of-the-art models are often trained with expensive online Reinforcement Learning (RL) or expert iteration. However, these approaches rely on fixed problem sets, which causes inefficient training and limits the model to tackle complex problems. To overcome these limitations, we propose GAR: Generative Adversarial Reinforcement learning, a comprehensive RL training framework that jointly trains the problem composer and solver in an adversarial loop. GAR introduces an implicit curriculum learning mechanism, which aligns task difficulty with the prover's evolving capability. It thereby improves the training efficiency and enables stronger performance of proving advanced theorems. Experiments show that with GAR training, Goedel-Prover-V2-8B and DeepSeek-Prover-V2-7B achieve an average relative improvement in pass@32 of 4.20% on MiniF2F-Test benchmark, while DeepSeek-Prover-V2's pass@32 on ProofNet-Test increases from 22.58% to 25.81%. Beyond formal proving, GAR establishes a general RL paradigm for co-evolution of problem generation and solving under verifiable environments.

A promising approach for improving reasoning in large language models is to use process reward models (PRMs). PRMs provide feedback at each step of a multi-step reasoning trace, potentially improving credit assignment over outcome reward models (ORMs) that only provide feedback at the final step. However, collecting dense, per-step human labels is not scalable, and training PRMs from automatically-labeled data has thus far led to limited gains. To improve a base policy by running search against a PRM or using it as dense rewards for reinforcement learning (RL), we ask: "How should we design process rewards?". Our key insight is that, to be effective, the process reward for a step should measure progress: a change in the likelihood of producing a correct response in the future, before and after taking the step, corresponding to the notion of step-level advantages in RL. Crucially, this progress should be measured under a prover policy distinct from the base policy. We theoretically characterize the set of good provers and our results show that optimizing process rewards from such provers improves exploration during test-time search and online RL. In fact, our characterization shows that weak prover policies can substantially improve a stronger base policy, which we also observe empirically. We validate our claims by training process advantage verifiers (PAVs) to predict progress under such provers, and show that compared to ORMs, test-time search against PAVs is >8% more accurate, and 1.5-5times more compute-efficient. Online RL with dense rewards from PAVs enables one of the first results with 5-6times gain in sample efficiency, and >6% gain in accuracy, over ORMs.

Formal methods is pivotal for verifying the reliability of critical systems through rigorous mathematical proofs. However, its adoption is hindered by labor-intensive manual proofs and the expertise required to use theorem provers. Recent advancements in large language models (LLMs) offer new opportunities for automated theorem proving. Two promising approaches are generating tactics step by step and generating a whole proof directly with an LLM. However, existing work makes no attempt to combine the two approaches. In this work, we introduce HybridProver, a dual-model proof synthesis framework that combines tactic-based generation and whole-proof synthesis to harness the benefits of both approaches. HybridProver generates whole proof candidates for evaluation directly, then extracts proof sketches from those candidates. It then uses a tactic-based generation model that integrates automated tools to complete the sketches via stepwise refinement. We implement HybridProver for the Isabelle theorem prover and fine-tune LLMs on our optimized Isabelle datasets. Evaluation on the miniF2F dataset illustrates HybridProver's effectiveness. We achieve a 59.4% success rate on miniF2F, where the previous SOTA is 56.1%. Our ablation studies show that this SOTA result is attributable to combining whole-proof and tactic-based generation. Additionally, we show how the dataset quality, training parameters, and sampling diversity affect the final result during automated theorem proving with LLMs. All of our code, datasets, and LLMs are open source.

We introduce SimpleQA Verified, a 1,000-prompt benchmark for evaluating Large Language Model (LLM) short-form factuality based on OpenAI's SimpleQA. It addresses critical limitations in OpenAI's benchmark, including noisy and incorrect labels, topical biases, and question redundancy. SimpleQA Verified was created through a rigorous multi-stage filtering process involving de-duplication, topic balancing, and source reconciliation to produce a more reliable and challenging evaluation set, alongside improvements in the autorater prompt. On this new benchmark, Gemini 2.5 Pro achieves a state-of-the-art F1-score of 55.6, outperforming other frontier models, including GPT-5. This work provides the research community with a higher-fidelity tool to track genuine progress in parametric model factuality and to mitigate hallucinations. The benchmark dataset, evaluation code, and leaderboard are available at: https://www.kaggle.com/benchmarks/deepmind/simpleqa-verified.

Translating human-written mathematical theorems and proofs from natural language (NL) into formal languages (FLs) like Lean 4 has long been a significant challenge for AI. Most state-of-the-art methods address this separately, first translating theorems and then generating proofs, creating a fundamental disconnect vis-a-vis true proof auto-formalization. This two-step process and its limitations were evident even in AlphaProof's silver-medal performance at the 2024 IMO, where problem statements needed manual translation before automated proof synthesis. We present ProofBridge, a unified framework for automatically translating entire NL theorems and proofs into Lean 4. At its core is a joint embedding model that aligns NL and FL (NL-FL) theorem-proof pairs in a shared semantic space, enabling cross-modal retrieval of semantically relevant FL examples to guide translation. Our training ensures that NL-FL theorems (and their proofs) are mapped close together in this space if and only if the NL-FL pairs are semantically equivalent. ProofBridge integrates retrieval-augmented fine-tuning with iterative proof repair, leveraging Lean's type checker and semantic equivalence feedback to ensure both syntactic correctness and semantic fidelity. Experiments show substantial improvements in proof auto-formalization over strong baselines (including GPT-5, Gemini-2.5, Kimina-Prover, DeepSeek-Prover), with our retrieval-augmented approach yielding significant gains in semantic correctness (SC, via proving bi-directional equivalence) and type correctness (TC, via type-checking theorem+proof) across pass@k metrics on miniF2F-Test-PF, a dataset we curated. In particular, ProofBridge improves cross-modal retrieval quality by up to 3.28x Recall@1 over all-MiniLM-L6-v2, and achieves +31.14% SC and +1.64% TC (pass@32) compared to the baseline Kimina-Prover-RL-1.7B.

The global economy is increasingly dependent on knowledge workers to meet the needs of public and private organizations. While there is no single definition of knowledge work, organizations and industry groups still attempt to measure individuals' capability to engage in it. The most comprehensive assessment of capability readiness for professional knowledge workers is the Uniform CPA Examination developed by the American Institute of Certified Public Accountants (AICPA). In this paper, we experimentally evaluate OpenAI's `text-davinci-003` and prior versions of GPT on both a sample Regulation (REG) exam and an assessment of over 200 multiple-choice questions based on the AICPA Blueprints for legal, financial, accounting, technology, and ethical tasks. First, we find that `text-davinci-003` achieves a correct rate of 14.4% on a sample REG exam section, significantly underperforming human capabilities on quantitative reasoning in zero-shot prompts. Second, `text-davinci-003` appears to be approaching human-level performance on the Remembering & Understanding and Application skill levels in the Exam absent calculation. For best prompt and parameters, the model answers 57.6% of questions correctly, significantly better than the 25% guessing rate, and its top two answers are correct 82.1% of the time, indicating strong non-entailment. Finally, we find that recent generations of GPT-3 demonstrate material improvements on this assessment, rising from 30% for `text-davinci-001` to 57% for `text-davinci-003`. These findings strongly suggest that large language models have the potential to transform the quality and efficiency of future knowledge work.

Autoformalization is the process of automatically translating from natural language mathematics to formal specifications and proofs. A successful autoformalization system could advance the fields of formal verification, program synthesis, and artificial intelligence. While the long-term goal of autoformalization seemed elusive for a long time, we show large language models provide new prospects towards this goal. We make the surprising observation that LLMs can correctly translate a significant portion (25.3%) of mathematical competition problems perfectly to formal specifications in Isabelle/HOL. We demonstrate the usefulness of this process by improving a previously introduced neural theorem prover via training on these autoformalized theorems. Our methodology results in a new state-of-the-art result on the MiniF2F theorem proving benchmark, improving the proof rate from 29.6% to 35.2%.

Proof assistants like Lean have revolutionized mathematical proof verification, ensuring high accuracy and reliability. Although large language models (LLMs) show promise in mathematical reasoning, their advancement in formal theorem proving is hindered by a lack of training data. To address this issue, we introduce an approach to generate extensive Lean 4 proof data derived from high-school and undergraduate-level mathematical competition problems. This approach involves translating natural language problems into formal statements, filtering out low-quality statements, and generating proofs to create synthetic data. After fine-tuning the DeepSeekMath 7B model on this synthetic dataset, which comprises 8 million formal statements with proofs, our model achieved whole-proof generation accuracies of 46.3% with 64 samples and 52% cumulatively on the Lean 4 miniF2F test, surpassing the baseline GPT-4 at 23.0% with 64 samples and a tree search reinforcement learning method at 41.0%. Additionally, our model successfully proved 5 out of 148 problems in the Lean 4 Formalized International Mathematical Olympiad (FIMO) benchmark, while GPT-4 failed to prove any. These results demonstrate the potential of leveraging large-scale synthetic data to enhance theorem-proving capabilities in LLMs. Both the synthetic dataset and the model will be made available to facilitate further research in this promising field.

Large Language Models (LLMs) should refuse to answer questions beyond their knowledge. This capability, which we term knowledge-aware refusal, is crucial for factual reliability. However, existing metrics fail to faithfully measure this ability. On the one hand, simple refusal-based metrics are biased by refusal rates and yield inconsistent scores when models exhibit different refusal tendencies. On the other hand, existing calibration metrics are proxy-based, capturing the performance of auxiliary calibration processes rather than the model's actual refusal behavior. In this work, we propose the Refusal Index (RI), a principled metric that measures how accurately LLMs refuse questions they do not know. We define RI as Spearman's rank correlation between refusal probability and error probability. To make RI practically measurable, we design a lightweight two-pass evaluation method that efficiently estimates RI from observed refusal rates across two standard evaluation runs. Extensive experiments across 16 models and 5 datasets demonstrate that RI accurately quantifies a model's intrinsic knowledge-aware refusal capability in factual tasks. Notably, RI remains stable across different refusal rates and provides consistent model rankings independent of a model's overall accuracy and refusal rates. More importantly, RI provides insight into an important but previously overlooked aspect of LLM factuality: while LLMs achieve high accuracy on factual tasks, their refusal behavior can be unreliable and fragile. This finding highlights the need to complement traditional accuracy metrics with the Refusal Index for comprehensive factuality evaluation.

Formal theorem proving with TLA+ provides rigorous guarantees for system specifications, but constructing proofs requires substantial expertise and effort. While large language models have shown promise in automating proofs for tactic-based theorem provers like Lean, applying these approaches directly to TLA+ faces significant challenges due to the hierarchical proof structure of the TLA+ proof system. We present a prompt-based approach that leverages LLMs to guide hierarchical decomposition of complex proof obligations into simpler sub-claims, while relying on symbolic provers for verification. Our key insight is to constrain LLMs to generate normalized claim decompositions rather than complete proofs, significantly reducing syntax errors. We also introduce a benchmark suite of 119 theorems adapted from (1) established mathematical collections and (2) inductive proofs of distributed protocols. Our approach consistently outperforms baseline methods across the benchmark suite.

Language models (LMs) are widely used by an increasing number of users, underscoring the challenge of maintaining factuality across a broad range of topics. We first present VERIFY (Verification and Evidence RetrIeval for FactualitY evaluation), a pipeline to evaluate LMs' factuality in real-world user interactions. VERIFY considers the verifiability of LM-generated content and categorizes content units as supported, unsupported, or undecidable based on the retrieved evidence from the Web. Importantly, factuality judgment by VERIFY correlates better with human evaluations than existing methods. Using VERIFY, we identify "hallucination prompts" across diverse topics, i.e., those eliciting the highest rates of incorrect and inconclusive LM responses. These prompts form FactBench, a dataset of 1K prompts across 150 fine-grained topics. Our dataset captures emerging factuality challenges in real-world LM interactions and can be regularly updated with new prompts. We benchmark widely-used LMs from GPT, Gemini, and Llama3.1 family on FactBench, yielding the following key findings: (i) Proprietary models exhibit better factuality, with performance declining from Easy to Hard hallucination prompts. (ii) Llama3.1-405B-Instruct shows comparable or lower factual accuracy than Llama3.1-70B-Instruct across all evaluation methods due to its higher subjectivity that leads to more content labeled as undecidable. (iii) Gemini1.5-Pro shows a significantly higher refusal rate, with over-refusal in 25% of cases. Our code and data are publicly available at https://huggingface.co/spaces/launch/factbench.

AI coding agents produce code faster than humans can review it. In our approach, the prover is the judge of whether the code is correct. Under a verifier-driven loop, AI agents wrote and verified bare-metal security software in Ada/SPARK spanning classical and post-quantum cryptography, TLS 1.3, IKEv2, X.509, and a Matrix client. GNATprove discharged 49,280 proof obligations, established functional correctness for selected primitives, and proved the absence of run-time errors for the rest, at roughly 20-40 times lower supervision cost than comparable hand verification. GNATprove alone was insufficient: some defects could not be detected and were resolved using known-answer tests, interoperability, or human review of specifications. Given weak checks, the agent tried to bypass them and reported success. We report where each layer caught faults and draw the central lesson: what an agent can be trusted to establish is bounded by the strength of its feedback.

Recent large language models (LLMs) have witnessed significant advancement in various tasks, including mathematical reasoning and theorem proving. As these two tasks require strict and formal multi-step inference, they are appealing domains for exploring the reasoning ability of LLMs but still face important challenges. Previous studies such as Chain-of-Thought (CoT) have revealed the effectiveness of intermediate steps guidance. However, such step-wise annotation requires heavy labor, leading to insufficient training steps for current benchmarks. To fill this gap, this work introduces MUSTARD, a data generation framework that masters uniform synthesis of theorem and proof data of high quality and diversity. MUSTARD synthesizes data in three stages: (1) It samples a few mathematical concept seeds as the problem category. (2) Then, it prompts a generative language model with the sampled concepts to obtain both the problems and their step-wise formal solutions. (3) Lastly, the framework utilizes a proof assistant (e.g., Lean Prover) to filter the valid proofs. With the proposed MUSTARD, we present a theorem-and-proof benchmark MUSTARDSAUCE with 5,866 valid data points. Each data point contains an informal statement, an informal proof, and a translated formal proof that passes the prover validation. We perform extensive analysis and demonstrate that MUSTARD generates validated high-quality step-by-step data. We further apply the MUSTARDSAUCE for fine-tuning smaller language models. The fine-tuned Llama 2-7B achieves a 15.41% average relative performance gain in automated theorem proving, and 8.18% in math word problems. Codes and data are available at https://github.com/Eleanor-H/MUSTARD.

Reinforcement Learning with Verifiable Rewards (RLVR) has become the standard paradigm for LLM mathematical reasoning, with Group Relative Policy Optimization (GRPO) serving as the dominant algorithm. We identify two overlooked inefficiencies inherent in GRPO. First, a fixed KL coefficient overly restricts policy exploration at moments when the model needs to diverge significantly from the reference policy. Second, uniform question sampling overlooks that moderately difficult problems produce the most informative gradient signals. We propose FG-ExPO, short for Frontier-Guided Exploration-Prioritized Policy Optimization, which integrates two lightweight components. Accuracy-Conditioned KL Scaling (AKL) adjusts the KL penalty strength through a smooth nonlinear function of batch average accuracy, loosening the constraint when the model performs poorly and strengthening it when the model achieves satisfactory results. Gaussian Curriculum Sampling (GCS) assigns sampling weights to questions following a Gaussian distribution centered at a moderate accuracy level around 0.5, focusing model training on its learning frontier. We conduct evaluations on DeepSeek-R1-Distill-Qwen-1.5B and Qwen3-8B-Base across six mainstream mathematical reasoning benchmarks. Experimental results demonstrate that FG-ExPO consistently outperforms vanilla GRPO. It delivers an absolute improvement of 13.34 on the AIME 2025 pass@32 metric, rising from 63.33 percent to 76.67 percent, and obtains an average pass@32 gain of 2.66 on the 8B model. The substantially larger performance gains observed on pass@32 compared to pass@1 verify that FG-ExPO enlarges the model's effective exploration space under a fixed inference budget.

Large language models for code generation increasingly rely on synthetic data, where both problem solutions and verification tests are generated by models. While this enables scalable data creation, it introduces a previously unexplored bottleneck: the verification ceiling, in which the quality and diversity of training data are fundamentally constrained by the capabilities of synthetic verifiers. In this work, we systematically study how verification design and strategies influence model performance. We investigate (i) what we verify by analyzing the impact of test complexity and quantity: richer test suites improve code generation capabilities (on average +3 pass@1), while quantity alone yields diminishing returns, (ii) how we verify by exploring relaxed pass thresholds: rigid 100% pass criteria can be overly restrictive. By allowing for relaxed thresholds or incorporating LLM-based soft verification, we can recover valuable training data, leading to a 2-4 point improvement in pass@1 performance. However, this benefit is contingent upon the strength and diversity of the test cases used, and (iii) why verification remains necessary through controlled comparisons of formally correct versus incorrect solutions and human evaluation: retaining diverse correct solutions per problem yields consistent generalization gains. Our results show that Verification as currently practiced is too rigid, filtering out valuable diversity. But it cannot be discarded, only recalibrated. By combining calibrated verification with diverse, challenging problem-solution pairs, we outline a path to break the verification ceiling and unlock stronger code generation models.

Agentic systems are evaluated on benchmarks where agents interact with environments to solve tasks. Most papers report a pass@1 score computed from a single run per task, assuming this gives a reliable performance estimate. We test this assumption by collecting 60,000 agentic trajectories on SWE-Bench-Verified, spanning three models and two scaffolds. We find substantial variance: single-run pass@1 estimates vary by 2.2 to 6.0 percentage points depending on which run is selected, with standard deviations exceeding 1.5 percentage points even at temperature 0. This variance has critical implications: reported improvements of 2--3 percentage points may reflect evaluation noise rather than genuine algorithmic progress. Through token-level analysis, we show that trajectories diverge early, often within the first few percent of tokens, and that these small differences cascade into different solution strategies. To enable reliable evaluation of agentic systems, we recommend three concrete practices: (1) estimate pass@1 from multiple independent runs per task, especially when measuring small improvements, (2) use statistical power analysis to determine the number of runs needed to detect expected effect sizes, and (3) consider metrics like pass@k (optimistic bound) and pass^k (pessimistic bound) with k>1 to better characterize the full performance envelope. While these practices increase evaluation cost, they are essential for distinguishing genuine scientific progress from statistical noise.

A fundamental challenge in formal theorem proving by LLMs is the lack of high-quality training data. Although reinforcement learning or expert iteration partially mitigates this issue by alternating between LLM generating proofs and finetuning them on correctly generated ones, performance quickly plateaus due to the scarcity of correct proofs (sparse rewards). To keep improving the models with limited data, we draw inspiration from mathematicians, who continuously develop new results, partly by proposing novel conjectures or exercises (which are often variants of known results) and attempting to solve them. We design the Self-play Theorem Prover (STP) that simultaneously takes on two roles, conjecturer and prover, each providing training signals to the other. The conjecturer is trained iteratively on previously generated conjectures that are barely provable by the current prover, which incentivizes it to generate increasingly challenging conjectures over time. The prover attempts to prove the conjectures with standard expert iteration. We evaluate STP with both Lean and Isabelle formal versifiers. With 19.8 billion tokens generated during the training in Lean, STP proves 26.3% of the statements in the LeanWorkbook dataset, doubling the previous best result of 13.2% achieved through expert iteration. The final model achieves state-of-the-art performance among whole-proof generation methods on miniF2F-test (61.7%, pass@3200), Proofnet-test (23.1%, pass@3200) and PutnamBench (8/644, pass@3200).

Large Language Models (LLMs) have made notable progress in automated theorem proving, yet existing formal benchmarks remain limited in both mathematical coverage and difficulty. Most are concentrated in areas that are easier to formalize, such as algebra and elementary number theory, and provide limited coverage of subfields that require deeper reasoning, including mathematical analysis. To address this gap, we introduce MA-ProofBench, to the best of our knowledge, the first formal theorem-proving benchmark dedicated to Mathematical Analysis. The benchmark contains 200 formalized theorems covering 6 core topics and 27 subcategories, including measure and integration theory, complex analysis, and functional analysis. The problems are divided into two difficulty levels, an undergraduate level (Level I, 100 problems) and a Ph.D. qualifying level (Level II, 100 problems), to evaluate how well LLMs perform formal reasoning at different mathematical depths. Each problem is constructed through a human-led, LLM-assisted formalization pipeline followed by independent expert review, ensuring that the formal statements remain faithful to the original mathematics. We evaluate a range of recent general-purpose reasoning models and formal theorem provers on MA-ProofBench. However, most models perform poorly: even the best-performing model, GPT-5.5, achieves only 16% Pass@8 on Level I and 5% on Level II, while most models stay close to 0% on Level II. Further analysis identifies Mathlib hallucinations and incomplete proofs as the two dominant failure modes, while an evaluation on the natural-language version of the benchmark exposes a clear gap between informal and formal reasoning. MA-ProofBench is intended to serve as a reliable reference for tracking progress in formal mathematical reasoning in advanced domains.

Reinforcement learning (RL) has become a central paradigm for post-training large language models (LLMs), particularly for complex reasoning tasks, yet it often suffers from exploration collapse: policies prematurely concentrate on a small set of dominant reasoning patterns, improving pass@1 while limiting rollout-level diversity and gains in pass@k. We argue that this failure stems from regularizing local token behavior rather than diversity over sets of solutions. To address this, we propose Uniqueness-Aware Reinforcement Learning, a rollout-level objective that explicitly rewards correct solutions that exhibit rare high-level strategies. Our method uses an LLM-based judge to cluster rollouts for the same problem according to their high-level solution strategies, ignoring superficial variations, and reweights policy advantages inversely with cluster size. As a result, correct but novel strategies receive higher rewards than redundant ones. Across mathematics, physics, and medical reasoning benchmarks, our approach consistently improves pass@k across large sampling budgets and increases the area under the pass@k curve (AUC@K) without sacrificing pass@1, while sustaining exploration and uncovering more diverse solution strategies at scale.

Corruption studies, the primary tool for evaluating chain-of-thought (CoT) faithfulness, identify which chain positions are "computationally important" by measuring accuracy when steps are replaced with errors. We identify a systematic confound: for chains with explicit terminal answer statements, the dominant format in standard benchmarks, corruption studies detect where the answer text appears, not where computation occurs. A within-dataset format ablation provides the key evidence: on standard GSM8K chains ending with "the answer is X," removing only the answer statement, preserving all reasoning, collapses suffix sensitivity ~19x at 3B (N=300, p=0.022). Conflicting-answer experiments quantify the causal mechanism: at 7B, CC accuracy drops to near-zero (<=0.02) across five architecture families; the followed-wrong rate spans 0.63-1.00 at 3B-7B and attenuates at larger scales (0.300 at Phi-4-14B, ~0.01 at 32B). A within-stable 7B replication (9.3x attenuation, N=76, p=7.8e-3; Qwen3-8B N=299, p=0.004) provides converging evidence, and the pattern replicates on MATH (DeepSeek-R1-7B: 10.9x suffix-survival recovery). On chains without answer suffixes the same protocol identifies the prefix as load-bearing (Delta=-0.77, p<10^-12). Generation-time probes confirm a dissociation: the answer is not early-determined during generation (early commitment <5%), yet at consumption time model outputs systematically follow the explicit answer text. The format-determination effect persists through 14B (8.5x ratio, p=0.001) and converges toward zero at 32B. We propose a three-prerequisite protocol (question-only control, format characterization, all-position sweep) as a minimum standard for corruption-based faithfulness studies.

State-of-the-art neural theorem provers like DeepSeek-Prover-V1.5 combine large language models with reinforcement learning, achieving impressive results through sophisticated training. We ask: do these highly-trained models still benefit from simple structural guidance at inference time? We evaluate a lightweight intervention -- a fixed prompt schedule over 15 common tactic skeletons -- on the miniF2F benchmark. This simple approach yields 21.7% pass@16 compared to 15.2% for standard sampling from the same model, a 43% relative improvement using the same number of samples (k=16) and same maximum generation length (1024 tokens). Our results suggest that even capable RL-trained provers underutilize structural priors available in the tactic language, and that simple inference-time guidance remains a cheap, complementary boost.

Commercial LLM services often conceal internal reasoning traces while still charging users for every generated token, including those from hidden intermediate steps, raising concerns of token inflation and potential overbilling. This gap underscores the urgent need for reliable token auditing, yet achieving it is far from straightforward: cryptographic verification (e.g., hash-based signature) offers little assurance when providers control the entire execution pipeline, while user-side prediction struggles with the inherent variance of reasoning LLMs, where token usage fluctuates across domains and prompt styles. To bridge this gap, we present PALACE (Predictive Auditing of LLM APIs via Reasoning Token Count Estimation), a user-side framework that estimates hidden reasoning token counts from prompt-answer pairs without access to internal traces. PALACE introduces a GRPO-augmented adaptation module with a lightweight domain router, enabling dynamic calibration across diverse reasoning tasks and mitigating variance in token usage patterns. Experiments on math, coding, medical, and general reasoning benchmarks show that PALACE achieves low relative error and strong prediction accuracy, supporting both fine-grained cost auditing and inflation detection. Taken together, PALACE represents an important first step toward standardized predictive auditing, offering a practical path to greater transparency, accountability, and user trust.

Despite rapid advances in large language models (LLMs), achieving reliable performance on highly professional and structured examinations remains a significant challenge. The Japanese bar examination is a particularly demanding benchmark, requiring not only advanced legal reasoning but also strict adherence to complex answer formats that involve joint evaluation of multiple propositions. While recent studies have reported improvements by decomposing such questions into simpler true--false judgments, these approaches have not been systematically evaluated under the original exam format and scoring scheme, leaving open the question of whether they truly capture exam-level competence. In this paper, we present a self-verification model trained on a newly constructed dataset that faithfully replicates the authentic format and evaluation scale of the exam. Our model is able to exceed the official passing score when evaluated on the actual exam scale, marking the first demonstration, to our knowledge, of an LLM passing the Japanese bar examination without altering its original question structure or scoring rules. We further conduct extensive comparisons with alternative strategies, including multi-agent inference and decomposition-based supervision, and find that these methods fail to achieve comparable performance. Our results highlight the importance of format-faithful supervision and consistency verification, and suggest that carefully designed single-model approaches can outperform more complex systems in high-stakes professional reasoning tasks. Our dataset and codes are publicly available.

Handwritten STEM exams capture open-ended reasoning and diagrams, but manual grading is slow and difficult to scale. We present an end-to-end workflow for grading scanned handwritten engineering quizzes with multimodal large language models (LLMs) that preserves the standard exam process (A4 paper, unconstrained student handwriting). The lecturer provides only a handwritten reference solution (100%) and a short set of grading rules; the reference is converted into a text-only summary that conditions grading without exposing the reference scan. Reliability is achieved through a multi-stage design with a format/presence check to prevent grading blank answers, an ensemble of independent graders, supervisor aggregation, and rigid templates with deterministic validation to produce auditable, machine-parseable reports. We evaluate the frozen pipeline in a clean-room protocol on a held-out real course quiz in Slovenian, including hand-drawn circuit schematics. With state-of-the-art backends (GPT-5.2 and Gemini-3 Pro), the full pipeline achieves approx8-point mean absolute difference to lecturer grades with low bias and an estimated manual-review trigger rate of approx17% at D_{max}=40. Ablations show that trivial prompting and removing the reference solution substantially degrade accuracy and introduce systematic over-grading, confirming that structured prompting and reference grounding are essential.

We introduce Kimina-Prover Preview, a large language model that pioneers a novel reasoning-driven exploration paradigm for formal theorem proving, as showcased in this preview release. Trained with a large-scale reinforcement learning pipeline from Qwen2.5-72B, Kimina-Prover demonstrates strong performance in Lean 4 proof generation by employing a structured reasoning pattern we term formal reasoning pattern. This approach allows the model to emulate human problem-solving strategies in Lean, iteratively generating and refining proof steps. Kimina-Prover sets a new state-of-the-art on the miniF2F benchmark, reaching 80.7% with pass@8192. Beyond improved benchmark performance, our work yields several key insights: (1) Kimina-Prover exhibits high sample efficiency, delivering strong results even with minimal sampling (pass@1) and scaling effectively with computational budget, stemming from its unique reasoning pattern and RL training; (2) we demonstrate clear performance scaling with model size, a trend previously unobserved for neural theorem provers in formal mathematics; (3) the learned reasoning style, distinct from traditional search algorithms, shows potential to bridge the gap between formal verification and informal mathematical intuition. We open source distilled versions with 1.5B and 7B parameters of Kimina-Prover

We introduce ProofNet, a benchmark for autoformalization and formal proving of undergraduate-level mathematics. The ProofNet benchmarks consists of 371 examples, each consisting of a formal theorem statement in Lean 3, a natural language theorem statement, and a natural language proof. The problems are primarily drawn from popular undergraduate pure mathematics textbooks and cover topics such as real and complex analysis, linear algebra, abstract algebra, and topology. We intend for ProofNet to be a challenging benchmark that will drive progress in autoformalization and automatic theorem proving. We report baseline results on statement autoformalization via in-context learning. Moreover, we introduce two novel statement autoformalization methods: prompt retrieval and distilled backtranslation.

Reinforcement learning with verifiable rewards has enabled strong post-training gains in domains such as math and coding, though many open-ended settings rely on rubric-based rewards. We study reward hacking in rubric-based RL, where a policy is optimized against a training verifier but evaluated against a cross-family panel of three frontier judges, reducing dependence on any single evaluator. Our framework separates two sources of divergence: verifier failure, where the training verifier credits rubric criteria that reference verifiers reject, and rubric-design limitations, where even strong rubric-based verifiers favor responses that rubric-free judges rate worse overall. Across medical and science domains, weak verifiers produce large proxy-reward gains that do not transfer to the reference verifiers; exploitation grows over training and concentrates in recurring failures such as partial satisfaction of compound criteria, treating implicit content as explicit, and imprecise topical matching. Stronger verifiers substantially reduce, but do not eliminate, verifier exploitation. We also introduce a self-internalization gap, a verifier-free diagnostic based on policy log-probabilities, which tracks reference-verifier quality, detecting when the policy trained using the weak verifier stops improving. Finally, in our setting, stronger verification does not prevent reward hacking when the rubric leaves important failure modes unspecified: rubric-based verifiers prefer the RL checkpoint, while rubric-free judges prefer the base model. These disagreements coincide with gains concentrated in completeness and presence-based criteria, alongside declines in factual correctness, conciseness, relevance, and overall quality. Together, these results suggest that stronger verification reduces reward hacking, but does not by itself ensure that rubric gains correspond to broader quality gains.

Frontier large language models are increasingly deployed as orchestration backbones for biological research workflows, yet no shared evidence base exists for comparing their refusal behaviour on legitimate research prompts. RefusalBench, introduced here, is a matched-triple benchmark of 141 prompts in 47 bundles that holds task framing constant while varying only biological risk tier (benign, borderline, dual-use), enabling tier-conditioned comparisons robust to subdomain confounding. A 15-prompt should-refuse positive-control module establishes per-model calibration floors; three models fail to refuse even these prompts. Across 19 frontier models in the May 2026 snapshot, strict refusal rates span 0.1% to 94.6% on identical prompts. Jurisdiction does not predict refusal in this snapshot (Mann-Whitney U, p = 0.393; EU n = 1, US bimodal); provider identity does, with Anthropic's API stack predicting refusal at OR = 21.03 (95% CI: 14.58-30.34 prompt-clustered; 5.70-77.55 under model-clustered GEE). This effect is best read as access-path-level rather than model-weight-level: 99.8% of Anthropic's strict refusals carry the same safety_policy adjudicated reason code, consistent with a small set of canonical refusal templates rather than case-by-case model reasoning. Strict refusal rate misranks safety calibration: Grok 4.20 achieves the highest tier discrimination (Youden's J = 0.787) while ranking only seventh by overall refusal rate, and Claude Opus 4.7's J dropped 65% from prior versions with no improvement in dual-use detection. Nine of 18 frontier models exhibit a hedge-but-help partial-compliance pattern at dual-use tier that binary refusal metrics cannot detect.

Large language models (LLMs) are entering legal workflows, yet we lack a jurisdiction-specific framework to assess their baseline competence therein. We use India's public legal examinations as a transparent proxy. Our multi-year benchmark assembles objective screens from top national and state exams and evaluates open and frontier LLMs under real-world exam conditions. To probe beyond multiple-choice questions, we also include a lawyer-graded, paired-blinded study of long-form answers from the Supreme Court's Advocate-on-Record exam. This is, to our knowledge, the first exam-grounded, India-specific yardstick for LLM court-readiness released with datasets and protocols. Our work shows that while frontier systems consistently clear historical cutoffs and often match or exceed recent top-scorer bands on objective exams, none surpasses the human topper on long-form reasoning. Grader notes converge on three reliability failure modes: procedural or format compliance, authority or citation discipline, and forum-appropriate voice and structure. These findings delineate where LLMs can assist (checks, cross-statute consistency, statute and precedent lookups) and where human leadership remains essential: forum-specific drafting and filing, procedural and relief strategy, reconciling authorities and exceptions, and ethical, accountable judgment.

The application of large language models (LLMs) in the field of coding is evolving rapidly: from code assistants, to autonomous coding agents, and then to generating complete projects through natural language. Early LLM code benchmarks primarily focused on code generation accuracy, but these benchmarks have gradually become saturated. Benchmark saturation weakens their guiding role for LLMs. For example, HumanEval Pass@1 has reached 99.4% and MBPP 94.2%. Among various attempts to address benchmark saturation, approaches based on software engineering have stood out, but the saturation of existing software engineering benchmarks is rapidly increasing. To address this, we propose a new benchmark, Web-Bench, which contains 50 projects, each consisting of 20 tasks with sequential dependencies. The tasks implement project features in sequence, simulating real-world human development workflows. When designing Web-Bench, we aim to cover the foundational elements of Web development: Web Standards and Web Frameworks. Given the scale and complexity of these projects, which were designed by engineers with 5 to 10 years of experience, each presents a significant challenge. On average, a single project takes 4 to 8 hours for a senior engineer to complete. On our given benchmark agent (Web-Agent), SOTA (Claude 3.7 Sonnet) achieves only 25.1% Pass@1, significantly lower (better) than SWE-Bench's Verified (65.4%) and Full (33.8%) scores. Finally, we discuss that in any development field, Standards and Frameworks represent foundational knowledge and efficiency tools, respectively, and LLMs require optimization tailored to them.

Many inference-time language-model pipelines combine a cheap reward signal with an expensive verifier, such as exact answer checking in mathematical reasoning or hidden-test execution in code generation. We formalize this setting using a learning-theoretic lens as generative active search: a cost-sensitive first-positive search problem in which a policy adaptively samples candidates from an unknown distribution, observes cheap scores, and pays for verifier labels until it finds a positive example. For a fixed prompt, the generator and reward model induce two unknown objects: a distribution over reward scores and a score-conditioned success function. When these quantities are known, we characterize the distribution-aware optimal policy using a dynamic programming approach. In the realistic and practical setting where both the score distribution and success function are unknown, we propose ADAP, a shellwise adaptive generate-rank-verify algorithm that progressively increases the number of sampled responses and top-ranked verifications. Under the monotonicity assumption that higher reward scores are no less likely to pass verification, we show that ADAP achieves expected cost within a constant factor of the distribution-aware optimum. We complement this result with learning-theoretic lower bounds, based on a centered star number, showing that structural assumptions on the score--label relationship are necessary. Experiments on mathematical reasoning and competitive programming validate the predicted advantage over both fixed non-adaptive policies and difficulty-adaptive baselines.

Mathematical reasoning has long served as a stringent test of machine intelligence; over the past decade, it has moved from a niche problem within NLP to one of the most consequential AI frontiers. This survey provides a unified account of the field's evolution, from early rule-based math word problem (MWP) solvers and template-driven geometry systems, through neural expression generation and LLM prompting, to contemporary reasoning models, multi-agent systems, neuro-symbolic theorem provers, and verified discovery workflows. We organize the landscape along four axes: (i) informal reasoning over text and diagrams, spanning MWP solving, multimodal geometry, and VLMs; (ii) formal reasoning in proof assistants, including autoformalization, tactic prediction, compiler-guided repair, and proof search; (iii) mathematical discovery, where systems propose constructions, improve bounds, or assist attacks on open problems; and (iv) the inference and training-time techniques, including CoT prompting, tool use, process reward models, and RLVR, that increasingly connect generation with verification. We catalog major benchmarks across grade-school arithmetic, competition mathematics, geometry, formal proving, multimodal and multilingual reasoning, and expert evaluation, and we examine benchmark saturation, contamination, reporting mismatches, and the distinction between pass@1, majority voting, and verifier-assisted pass@k. We critically assess failure modes: brittleness under perturbation, reward hacking, multimodal grounding failures, fragile formalization, and the energy cost of reasoning-scale inference. Drawing on recent perspectives from working mathematicians, we identify future directions centered on verified-discovery workflows, reasoning efficiency, and infrastructure to make AI-assisted formalization broadly usable. Companion materials: https://github.com/Starscream-11813/awesome-AI4Math.

A paraphrase-quality audit of MathCheck (ICLR 2025) detected 4 semantically incorrect paraphrases in 129 groups (3.1%); removing them drops GPT-4o from rank 2 to rank 4 and elevates Claude Haiku and DeepSeek V3 above it; these ranking changes are invisible to any single-model evaluation. Cross-model unanimity found these errors automatically (>= 3/4 models for MathCheck; >= 6/9 for our primary evaluation) for under $10; in our own dataset the same protocol found that 47% of auto-generated connective-variation paraphrases were semantically incorrect. That flaw compounds a deeper measurement gap: Claude Haiku 4.5 achieves 86% accuracy yet SCR=50%, meaning half its theorems are answered differently under semantically equivalent restatements, while aggregate accuracy across 9 models spans only 86-96% yet Semantic Consistency Rates (SCR) span 50-82% -- a 32-point gap invisible to standard benchmarks. Formally, for any target ranking over 9 frontier models there exists a weighting over paraphrase families that realizes it (No-Free-Benchmark corollary), because no model Pareto-dominates all families -- so benchmark designers who select families are implicitly choosing which model wins. FormInv supplies the audit protocol (replicated on external benchmarks at 100% recall), SCR and per-theorem Cochran's Q as primary invariance measures evaluated on 9 models across 366-811 items (on Lean4-verified theorems), and FormInvSelector for regime-aware model selection.

Large language models (LLMs) have demonstrated solid zero-shot reasoning capabilities, which is reflected in their performance on the current test tasks. This calls for a more challenging benchmark requiring highly advanced reasoning ability to be solved. In this paper, we introduce such a benchmark, consisting of 191 long-form (1200 words on average) mystery narratives constructed as detective puzzles. Puzzles are sourced from the "5 Minute Mystery" platform and include a multiple-choice question for evaluation. Only 47% of humans solve a puzzle successfully on average, while the best human solvers achieve over 80% success rate. We show that GPT-3 models barely outperform random on this benchmark (with 28% accuracy) while state-of-the-art GPT-4 solves only 38% of puzzles. This indicates that there is still a significant gap in the deep reasoning abilities of LLMs and humans and highlights the need for further research in this area. Our work introduces a challenging benchmark for future studies on reasoning in language models and contributes to a better understanding of the limits of LLMs' abilities.