Daily Papers

The rapid evolution of Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) has produced substantial gains in reasoning, perception, and generative capability across language and vision. However, whether these advances yield commensurate improvements in safety remains unclear, in part due to fragmented evaluation practices limited to single modalities or threat models. In this report, we present an integrated safety evaluation of 7 frontier models: GPT-5.2, Gemini 3 Pro, Qwen3-VL, Doubao 1.8, Grok 4.1 Fast, Nano Banana Pro, and Seedream 4.5. We evaluate each model across language, vision-language, and image generation settings using a unified protocol that integrates benchmark evaluation, adversarial evaluation, multilingual evaluation, and compliance evaluation. Aggregating our evaluations into safety leaderboards and model safety profiles across multiple evaluation modes reveals a sharply heterogeneous safety landscape. While GPT-5.2 demonstrates consistently strong and balanced safety performance across evaluations, other models exhibit pronounced trade-offs among benchmark safety, adversarial alignment, multilingual generalization, and regulatory compliance. Both language and vision-language modalities show significant vulnerability under adversarial evaluation, with all models degrading substantially despite strong results on standard benchmarks. Text-to-image models achieve relatively stronger alignment in regulated visual risk categories, yet remain brittle under adversarial or semantically ambiguous prompts. Overall, these results show that safety in frontier models is inherently multidimensional--shaped by modality, language, and evaluation scheme, underscoring the need for standardized safety evaluations to accurately assess real-world risk and guide responsible model development and deployment.

We propose Zero-Error Horizon (ZEH) for trustworthy LLMs, which represents the maximum range that a model can solve without any errors. While ZEH itself is simple, we demonstrate that evaluating the ZEH of state-of-the-art LLMs yields abundant insights. For example, by evaluating the ZEH of GPT-5.2, we found that GPT-5.2 cannot even compute the parity of a short string like 11000, and GPT-5.2 cannot determine whether the parentheses in ((((()))))) are balanced. This is surprising given the excellent capabilities of GPT-5.2. The fact that LLMs make mistakes on such simple problems serves as an important lesson when applying LLMs to safety-critical domains. By applying ZEH to Qwen2.5 and conducting detailed analysis, we found that while ZEH correlates with accuracy, the detailed behaviors differ, and ZEH provides clues about the emergence of algorithmic capabilities. Finally, while computing ZEH incurs significant computational cost, we discuss how to mitigate this cost by achieving up to one order of magnitude speedup using tree structures and online softmax.

AI agents are expected to perform professional work across hundreds of occupational domains (from emergency department triage to nuclear reactor safety monitoring to customs import processing), yet existing benchmarks can only evaluate agents in the few domains where public environments exist. We introduce OccuBench, a benchmark covering 100 real-world professional task scenarios across 10 industry categories and 65 specialized domains, enabled by Language World Models (LWMs) that simulate domain-specific environments through LLM-driven tool response generation. Our multi-agent synthesis pipeline automatically produces evaluation instances with guaranteed solvability, calibrated difficulty, and document-grounded diversity. OccuBench evaluates agents along two complementary dimensions: task completion across professional domains and environmental robustness under controlled fault injection (explicit errors, implicit data degradation, and mixed faults). We evaluate 15 frontier models across 8 model families and find that: (1) no single model dominates all industries, as each has a distinct occupational capability profile; (2) implicit faults (truncated data, missing fields) are harder than both explicit errors (timeouts, 500s) and mixed faults, because they lack overt error signals and require the agent to independently detect data degradation; (3) larger models, newer generations, and higher reasoning effort consistently improve performance. GPT-5.2 improves by 27.5 points from minimal to maximum reasoning effort; and (4) strong agents are not necessarily strong environment simulators. Simulator quality is critical for LWM-based evaluation reliability. OccuBench provides the first systematic cross-industry evaluation of AI agents on professional occupational tasks.

Self-consistency has emerged as a popular technique for improving large language model accuracy on reasoning tasks. The approach is straightforward: generate multiple reasoning paths and select the most common answer through majority voting. While this reliably boosts accuracy, it remains unclear whether these gains reflect genuine improvements in reasoning quality. We investigate a fundamental question that has not been studied before: does inference scaling improve reasoning faithfulness? We conduct a comprehensive empirical study across four frontier models (GPT-5.2, Claude Opus 4.5, Gemini-3-flash-preview, and DeepSeek-v3.2) on 100 GSM8K mathematical reasoning problems. Our analysis employs bootstrap confidence intervals, McNemar's tests for paired comparisons, and Cohen's d effect sizes to quantify the effects rigorously. The results reveal striking differences across models that challenge common assumptions about self-consistency. GPT-5.2 shows the expected pattern: accuracy improves from 78% to 90% at N=5, with faithfulness remaining relatively stable (0.540 to 0.510). Claude Opus 4.5 tells a completely different story. Its accuracy actually drops from 78% to 74.3% while faithfulness jumps dramatically from 0.270 to 0.891 at N=5. DeepSeek-v3.2, already at 98% accuracy, shows ceiling effects with modest faithfulness gains (0.440 to 0.541). Gemini-3-flash improves from 81% to 86% accuracy with a slight faithfulness decrease (0.260 to 0.212). Problem difficulty analysis reveals that GPT-5.2 solves 82% of hard problems while breaking only 13% of easy ones. Claude, in contrast, breaks 23% of easy problems, explaining its accuracy decrease. These findings matter for practitioners: self-consistency is not universally beneficial, and teams should test their specific models before deployment. We release our code and provide practical recommendations for navigating these tradeoffs.

Financial markets are noisy and non-stationary, making alpha mining highly sensitive to noise in backtesting results and sudden market regime shifts. While recent agentic frameworks improve alpha mining automation, they often lack controllable multi-round search and reliable reuse of validated experience. To address these challenges, we propose QuantaAlpha, an evolutionary alpha mining framework that treats each end-to-end mining run as a trajectory and improves factors through trajectory-level mutation and crossover operations. QuantaAlpha localizes suboptimal steps in each trajectory for targeted revision and recombines complementary high-reward segments to reuse effective patterns, enabling structured exploration and refinement across mining iterations. During factor generation, QuantaAlpha enforces semantic consistency across the hypothesis, factor expression, and executable code, while constraining the complexity and redundancy of the generated factor to mitigate crowding. Extensive experiments on the China Securities Index 300 (CSI 300) demonstrate consistent gains over strong baseline models and prior agentic systems. When utilizing GPT-5.2, QuantaAlpha achieves an Information Coefficient (IC) of 0.1501, with an Annualized Rate of Return (ARR) of 27.75% and a Maximum Drawdown (MDD) of 7.98%. Moreover, factors mined on CSI 300 transfer effectively to the China Securities Index 500 (CSI 500) and the Standard & Poor's 500 Index (S&P 500), delivering 160% and 137% cumulative excess return over four years, respectively, which indicates strong robustness of QuantaAlpha under market distribution shifts.

Background: Reporting and Data Systems (RADS) standardize radiology risk communication but automated RADS assignment from narrative reports is challenging because of guideline complexity, output-format constraints, and limited benchmarking across RADS frameworks and model sizes. Purpose: To create RXL-RADSet, a radiologist-verified synthetic multi-RADS benchmark, and compare validity and accuracy of open-weight small language models (SLMs) with a proprietary model for RADS assignment. Materials and Methods: RXL-RADSet contains 1,600 synthetic radiology reports across 10 RADS (BI-RADS, CAD-RADS, GB-RADS, LI-RADS, Lung-RADS, NI-RADS, O-RADS, PI-RADS, TI-RADS, VI-RADS) and multiple modalities. Reports were generated by LLMs using scenario plans and simulated radiologist styles and underwent two-stage radiologist verification. We evaluated 41 quantized SLMs (12 families, 0.135-32B parameters) and GPT-5.2 under a fixed guided prompt. Primary endpoints were validity and accuracy; a secondary analysis compared guided versus zero-shot prompting. Results: Under guided prompting GPT-5.2 achieved 99.8% validity and 81.1% accuracy (1,600 predictions). Pooled SLMs (65,600 predictions) achieved 96.8% validity and 61.1% accuracy; top SLMs in the 20-32B range reached ~99% validity and mid-to-high 70% accuracy. Performance scaled with model size (inflection between <1B and >=10B) and declined with RADS complexity primarily due to classification difficulty rather than invalid outputs. Guided prompting improved validity (99.2% vs 96.7%) and accuracy (78.5% vs 69.6%) compared with zero-shot. Conclusion: RXL-RADSet provides a radiologist-verified multi-RADS benchmark; large SLMs (20-32B) can approach proprietary-model performance under guided prompting, but gaps remain for higher-complexity schemes.

We show that training AI agents on high-fidelity reinforcement learning environments produces capabilities that generalize beyond the training distribution. We introduce CoreCraft, the first environment in EnterpriseBench, Surge AI's suite of agentic RL environments. CoreCraft is a fully operational enterprise simulation of a customer support organization, comprising over 2,500 entities across 14 entity types with 23 unique tools, designed to measure whether AI agents can perform the multi-step, domain-specific work that real jobs demand. Frontier models such as GPT-5.2 and Claude Opus 4.6 solve fewer than 30% of tasks when all expert-authored rubric criteria must be satisfied. Using this environment, we train GLM 4.6 with Group Relative Policy Optimization (GRPO) and adaptive clipping. After a single epoch of training, the model improves from 25.37% to 36.76% task pass rate on held-out evaluation tasks. More importantly, these gains transfer to out-of-distribution benchmarks: +4.5% on BFCL Parallel, +7.4% on Tau2-Bench Retail, and +6.8% on Tool Decathlon (Pass@1). We believe three environment properties are consistent with the observed transfer: task-centric world building that optimizes for diverse, challenging tasks; expert-authored rubrics enabling reliable reward computation; and enterprise workflows that reflect realistic professional patterns. Our results suggest that environment quality, diversity, and realism are key factors enabling generalizable agent capabilities.

Automatic evaluation of large language model (LLM) responses requires not only factual correctness but also clarity, particularly in political question-answering. While recent datasets provide human annotations for clarity and evasion, the impact of prompt design on automatic clarity evaluation remains underexplored. In this paper, we study prompt-based clarity evaluation using the CLARITY dataset from the SemEval 2026 shared task. We compare a GPT-3.5 baseline provided with the dataset against GPT-5.2 evaluated under three prompting strategies: simple prompting, chain-of-thought prompting, and chain-of-thought with few-shot examples. Model predictions are evaluated against human annotations using accuracy and class-wise metrics for clarity and evasion, along with hierarchical exact match. Results show that GPT-5.2 consistently outperforms the GPT-3.5 baseline on clarity prediction, with accuracy improving from 56 percent to 63 percent under chain-of-thought with few-shot prompting. Chain-of-thought prompting yields the highest evasion accuracy at 34 percent, though improvements are less stable across fine-grained evasion categories. We further evaluate topic identification and find that reasoning-based prompting improves accuracy from 60 percent to 74 percent relative to human annotations. Overall, our findings indicate that prompt design reliably improves high-level clarity evaluation, while fine-grained evasion and topic detection remain challenging despite structured reasoning prompts.

We present a system for training enterprise search agents via reinforcement learning that achieves state-of-the-art performance across a diverse suite of hard-to-verify agentic search tasks. Our work makes four core contributions. First, we introduce KARLBench, a multi-capability evaluation suite spanning six distinct search regimes, including constraint-driven entity search, cross-document report synthesis, tabular numerical reasoning, exhaustive entity retrieval, procedural reasoning over technical documentation, and fact aggregation over internal enterprise notes. Second, we show that models trained across heterogeneous search behaviors generalize substantially better than those optimized for any single benchmark. Third, we develop an agentic synthesis pipeline that employs long-horizon reasoning and tool use to generate diverse, grounded, and high-quality training data, with iterative bootstrapping from increasingly capable models. Fourth, we propose a new post-training paradigm based on iterative large-batch off-policy RL that is sample efficient, robust to train-inference engine discrepancies, and naturally extends to multi-task training with out-of-distribution generalization. Compared to Claude 4.6 and GPT 5.2, KARL is Pareto-optimal on KARLBench across cost-quality and latency-quality trade-offs, including tasks that were out-of-distribution during training. With sufficient test-time compute, it surpasses the strongest closed models. These results show that tailored synthetic data in combination with multi-task reinforcement learning enables cost-efficient and high-performing knowledge agents for grounded reasoning.

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.

The rapid integration of Multimodal Large Language Models (MLLMs) into critical applications is increasingly hindered by persistent safety vulnerabilities. However, existing red-teaming benchmarks are often fragmented, limited to single-turn text interactions, and lack the scalability required for systematic evaluation. To address this, we introduce OpenRT, a unified, modular, and high-throughput red-teaming framework designed for comprehensive MLLM safety evaluation. At its core, OpenRT architects a paradigm shift in automated red-teaming by introducing an adversarial kernel that enables modular separation across five critical dimensions: model integration, dataset management, attack strategies, judging methods, and evaluation metrics. By standardizing attack interfaces, it decouples adversarial logic from a high-throughput asynchronous runtime, enabling systematic scaling across diverse models. Our framework integrates 37 diverse attack methodologies, spanning white-box gradients, multi-modal perturbations, and sophisticated multi-agent evolutionary strategies. Through an extensive empirical study on 20 advanced models (including GPT-5.2, Claude 4.5, and Gemini 3 Pro), we expose critical safety gaps: even frontier models fail to generalize across attack paradigms, with leading models exhibiting average Attack Success Rates as high as 49.14%. Notably, our findings reveal that reasoning models do not inherently possess superior robustness against complex, multi-turn jailbreaks. By open-sourcing OpenRT, we provide a sustainable, extensible, and continuously maintained infrastructure that accelerates the development and standardization of AI safety.

As language models shift from single-shot answer generation toward multi-step reasoning that retrieves and consumes evidence mid-inference, evaluating the role of individual retrieved items becomes more important. Existing RAG evaluation typically targets final-answer quality, citation faithfulness, or answer-level attribution, but none of these directly targets the intervention-based, per-evidence-item utility view we study here. We introduce CUE-R, a lightweight intervention-based framework for measuring per-evidence-item operational utility in single-shot RAG using shallow observable retrieval-use traces. CUE-R perturbs individual evidence items via REMOVE, REPLACE, and DUPLICATE operators, then measures changes along three utility axes (correctness, proxy-based grounding faithfulness, and confidence error) plus a trace-divergence signal. We also outline an operational evidence-role taxonomy for interpreting intervention outcomes. Experiments on HotpotQA and 2WikiMultihopQA with Qwen-3 8B and GPT-5.2 reveal a consistent pattern: REMOVE and REPLACE substantially harm correctness and grounding while producing large trace shifts, whereas DUPLICATE is often answer-redundant yet not fully behaviorally neutral. A zero-retrieval control confirms that these effects arise from degradation of meaningful retrieval. A two-support ablation further shows that multi-hop evidence items can interact non-additively: removing both supports harms performance far more than either single removal. Our results suggest that answer-only evaluation misses important evidence effects and that intervention-based utility analysis is a practical complement for RAG evaluation.

Large Language Model (LLM) code agents increasingly resolve repository-level issues by iteratively editing code, invoking tools, and validating candidate patches. In these workflows, agents often write tests on the fly, a paradigm adopted by many high-ranking agents on the SWE-bench leaderboard. However, we observe that GPT-5.2, which writes almost no new tests, can even achieve performance comparable to top-ranking agents. This raises the critical question: whether such tests meaningfully improve issue resolution or merely mimic human testing practices while consuming a substantial interaction budget. To reveal the impact of agent-written tests, we present an empirical study that analyzes agent trajectories across six state-of-the-art LLMs on SWE-bench Verified. Our results show that while test writing is commonly adopted, but resolved and unresolved tasks within the same model exhibit similar test-writing frequencies Furthermore, these tests typically serve as observational feedback channels, where agents prefer value-revealing print statements significantly more than formal assertion-based checks. Based on these insights, we perform a controlled experiment by revising the prompts of four agents to either increase or reduce test writing. The results suggest that changes in the volume of agent-written tests do not significantly change final outcomes. Taken together, our study reveals that current test-writing practices may provide marginal utility in autonomous software engineering tasks.

We introduce ResearchGym, a benchmark and execution environment for evaluating AI agents on end-to-end research. To instantiate this, we repurpose five oral and spotlight papers from ICML, ICLR, and ACL. From each paper's repository, we preserve the datasets, evaluation harness, and baseline implementations but withhold the paper's proposed method. This results in five containerized task environments comprising 39 sub-tasks in total. Within each environment, agents must propose novel hypotheses, run experiments, and attempt to surpass strong human baselines on the paper's metrics. In a controlled evaluation of an agent powered by GPT-5, we observe a sharp capability--reliability gap. The agent improves over the provided baselines from the repository in just 1 of 15 evaluations (6.7%) by 11.5%, and completes only 26.5% of sub-tasks on average. We identify recurring long-horizon failure modes, including impatience, poor time and resource management, overconfidence in weak hypotheses, difficulty coordinating parallel experiments, and hard limits from context length. Yet in a single run, the agent surpasses the solution of an ICML 2025 Spotlight task, indicating that frontier agents can occasionally reach state-of-the-art performance, but do so unreliably. We additionally evaluate proprietary agent scaffolds including Claude Code (Opus-4.5) and Codex (GPT-5.2) which display a similar gap. ResearchGym provides infrastructure for systematic evaluation and analysis of autonomous agents on closed-loop research.

Large language models are increasingly used as agents in social, economic, and policy simulations. A common assumption is that stronger reasoning should improve simulation fidelity. We argue that this assumption can fail when the objective is not to solve a strategic problem, but to sample plausible boundedly rational behavior. In such settings, reasoning-enhanced models can become better solvers and worse simulators: they can over-optimize for strategically dominant actions, collapse compromise-oriented terminal behavior, and sometimes exhibit a diversity-without-fidelity pattern in which local variation survives without outcome-level fidelity. We study this solver-sampler mismatch in three multi-agent negotiation environments adapted from earlier simulation work: an ambiguous fragmented-authority trading-limits scenario, an ambiguous unified-opposition trading-limits scenario, and a new-domain grid-curtailment case in emergency electricity management. We compare three reflection conditions, no reflection, bounded reflection, and native reasoning, across two primary model families and then extend the same protocol to direct OpenAI runs with GPT-4.1 and GPT-5.2. Across all three experiments, bounded reflection produces substantially more diverse and compromise-oriented trajectories than either no reflection or native reasoning. In the direct OpenAI extension, GPT-5.2 native ends in authority decisions in 45 of 45 runs across the three experiments, while GPT-5.2 bounded recovers compromise outcomes in every environment. The contribution is not a claim that reasoning is generally harmful. It is a methodological warning: model capability and simulation fidelity are different objectives, and behavioral simulation should qualify models as samplers, not only as solvers.

Despite significant strides in language models in the last few years, when used as agents, such models often try to perform actions that are not just suboptimal for a given state, but are strictly prohibited by the external environment. For example, in the recent Kaggle GameArena chess competition, 78% of Gemini-2.5-Flash losses were attributed to illegal moves. Often people manually write "harnesses" around LLMs to prevent such failures. In this paper, we demonstrate that Gemini-2.5-Flash can automatically synthesize such a code harness, using a small number of rounds of iterative code refinement given feedback from the (game) environment. The resulting harness prevents all illegal moves in 145 different TextArena games (both 1-player and 2-player), enabling the smaller Gemini-2.5-Flash model to outperform larger models, such as Gemini-2.5-Pro. Pushing our technique to the limit, we can get Gemini-2.5-Flash to generate the entire policy in code, thus eliminating the need to use the LLM at decision making time. The resulting code-policy receives a higher average reward than Gemini-2.5-Pro and GPT-5.2-High on 16 TextArena 1-player games. Our results show that using a smaller model to synthesize a custom code harness (or entire policy) can outperform a much larger model, while also being more cost effective.

Great scientists have strong judgement and foresight, closely tied to what we call scientific taste. Here, we use the term to refer to the capacity to judge and propose research ideas with high potential impact. However, most relative research focuses on improving an AI scientist's executive capability, while enhancing an AI's scientific taste remains underexplored. In this work, we propose Reinforcement Learning from Community Feedback (RLCF), a training paradigm that uses large-scale community signals as supervision, and formulate scientific taste learning as a preference modeling and alignment problem. For preference modeling, we train Scientific Judge on 700K field- and time-matched pairs of high- vs. low-citation papers to judge ideas. For preference alignment, using Scientific Judge as a reward model, we train a policy model, Scientific Thinker, to propose research ideas with high potential impact. Experiments show Scientific Judge outperforms SOTA LLMs (e.g., GPT-5.2, Gemini 3 Pro) and generalizes to future-year test, unseen fields, and peer-review preference. Furthermore, Scientific Thinker proposes research ideas with higher potential impact than baselines. Our findings show that AI can learn scientific taste, marking a key step toward reaching human-level AI scientists.

As language models are increasingly deployed for complex autonomous tasks, their ability to reason accurately over longer horizons becomes critical. An essential component of this ability is planning and managing a long, complex chain-of-thought (CoT). We introduce LongCoT, a scalable benchmark of 2,500 expert-designed problems spanning chemistry, mathematics, computer science, chess, and logic to isolate and directly measure the long-horizon CoT reasoning capabilities of frontier models. Problems consist of a short input with a verifiable answer; solving them requires navigating a graph of interdependent steps that span tens to hundreds of thousands of reasoning tokens. Each local step is individually tractable for frontier models, so failures reflect long-horizon reasoning limitations. At release, the best models achieve <10% accuracy (GPT 5.2: 9.8%; Gemini 3 Pro: 6.1%) on LongCoT, revealing a substantial gap in current capabilities. Overall, LongCoT provides a rigorous measure of long-horizon reasoning, tracking the ability of frontier models to reason reliably over extended periods.

A well-calibrated model should express confidence that matches its actual accuracy -- when it claims 80\% confidence, it should be correct 80\% of the time. While large language models (LLMs) have achieved remarkable performance across diverse tasks, their epistemic calibration remains poorly understood. We introduce KalshiBench, a benchmark of 300 prediction market questions from Kalshi, a CFTC-regulated exchange, with verifiable real-world outcomes occurring after model training cutoffs. Unlike traditional benchmarks measuring accuracy on static knowledge, KalshiBench evaluates whether models can appropriately quantify uncertainty about genuinely unknown future events. We evaluate five frontier models -- Claude Opus 4.5, GPT-5.2, DeepSeek-V3.2, Qwen3-235B, and Kimi-K2 -- and find systematic overconfidence across all models. Even the best-calibrated model (Claude Opus 4.5, ECE=0.120) shows substantial calibration errors, while reasoning-enhanced models like GPT-5.2-XHigh exhibit worse calibration (ECE=0.395) despite comparable accuracy. Critically, only one model achieves a positive Brier Skill Score, indicating most models perform worse than simply predicting base rates. Our findings suggest that scaling and enhanced reasoning do not automatically confer calibration benefits, highlighting epistemic calibration as a distinct capability requiring targeted development.

Scalable Vector Graphics (SVG) are an essential format for technical illustration and digital design, offering precise resolution independence and flexible semantic editability. In practice, however, original vector source files are frequently lost or inaccessible, leaving only "flat" rasterized versions (e.g., PNG or JPEG) that are difficult to modify or scale. Manually reconstructing these figures is a prohibitively labor-intensive process, requiring specialized expertise to recover the original geometric intent. To bridge this gap, we propose VFIG, a family of Vision-Language Models trained for complex and high-fidelity figure-to-SVG conversion. While this task is inherently data-driven, existing datasets are typically small-scale and lack the complexity of professional diagrams. We address this by introducing VFIG-DATA, a large-scale dataset of 66K high-quality figure-SVG pairs, curated from a diverse mix of real-world paper figures and procedurally generated diagrams. Recognizing that SVGs are composed of recurring primitives and hierarchical local structures, we introduce a coarse-to-fine training curriculum that begins with supervised fine-tuning (SFT) to learn atomic primitives and transitions to reinforcement learning (RL) refinement to optimize global diagram fidelity, layout consistency, and topological edge cases. Finally, we introduce VFIG-BENCH, a comprehensive evaluation suite with novel metrics designed to measure the structural integrity of complex figures. VFIG achieves state-of-the-art performance among open-source models and performs on par with GPT-5.2, achieving a VLM-Judge score of 0.829 on VFIG-BENCH.

LLMs are increasingly used for code generation, but their outputs often follow recurring templates that can induce predictable vulnerabilities. We study vulnerability persistence in LLM-generated software and introduce Feature--Security Table (FSTab) with two components. First, FSTab enables a black-box attack that predicts likely backend vulnerabilities from observable frontend features and knowledge of the source LLM, without access to backend code or source code. Second, FSTab provides a model-centric evaluation that quantifies how consistently a given model reproduces the same vulnerabilities across programs, semantics-preserving rephrasings, and application domains. We evaluate FSTab on state-of-the-art code LLMs, including GPT-5.2, Claude-4.5 Opus, and Gemini-3 Pro, across diverse application domains. Our results show strong cross-domain transfer: even when the target domain is excluded from training, FSTab achieves up to 94\% attack success and 93\% vulnerability coverage on Internal Tools (Claude-4.5 Opus). These findings expose an underexplored attack surface in LLM-generated software and highlight the security risks of code generation. Our code is available at: https://anonymous.4open.science/r/FSTab-024E.

Developers and consumers increasingly choose reasoning language models (RLMs) based on their listed API prices. However, how accurately do these prices reflect actual inference costs? We conduct the first systematic study of this question, evaluating 8 frontier RLMs across 9 diverse tasks covering competition math, science QA, code generation, and multi-domain reasoning. We uncover the pricing reversal phenomenon: in 21.8% of model-pair comparisons, the model with a lower listed price actually incurs a higher total cost, with reversal magnitude reaching up to 28x. For example, Gemini 3 Flash's listed price is 78% cheaper than GPT-5.2's, yet its actual cost across all tasks is 22% higher. We trace the root cause to vast heterogeneity in thinking token consumption: on the same query, one model may use 900% more thinking tokens than another. In fact, removing thinking token costs reduces ranking reversals by 70% and raises the rank correlation (Kendall's τ ) between price and cost rankings from 0.563 to 0.873. We further show that per-query cost prediction is fundamentally difficult: repeated runs of the same query yield thinking token variation up to 9.7x, establishing an irreducible noise floor for any predictor. Our findings demonstrate that listed API pricing is an unreliable proxy for actual cost, calling for cost-aware model selection and transparent per-request cost monitoring.

We introduce the AI Productivity Index for Agents (APEX-Agents), a benchmark for assessing whether AI agents can execute long-horizon, cross-application tasks created by investment banking analysts, management consultants, and corporate lawyers. APEX-Agents requires agents to navigate realistic work environments with files and tools. We test eight agents for the leaderboard using Pass@1. Gemini 3 Flash (Thinking=High) achieves the highest score of 24.0%, followed by GPT-5.2 (Thinking=High), Claude Opus 4.5 (Thinking=High), and Gemini 3 Pro (Thinking=High). We open source the APEX-Agents benchmark (n=480) with all prompts, rubrics, gold outputs, files, and metadata. We also open-source Archipelago, our infrastructure for agent execution and evaluation.

Recent LLM-based data agents aim to automate data science tasks ranging from data analysis to deep learning. However, the open-ended nature of real-world data science problems, which often span multiple taxonomies and lack standard answers, poses a significant challenge for evaluation. To address this, we introduce DSAEval, a benchmark comprising 641 real-world data science problems grounded in 285 diverse datasets, covering both structured and unstructured data (e.g., vision and text). DSAEval incorporates three distinctive features: (1) Multimodal Environment Perception, which enables agents to interpret observations from multiple modalities including text and vision; (2) Multi-Query Interactions, which mirror the iterative and cumulative nature of real-world data science projects; and (3) Multi-Dimensional Evaluation, which provides a holistic assessment across reasoning, code, and results. We systematically evaluate 11 advanced agentic LLMs using DSAEval. Our results show that Claude-Sonnet-4.5 achieves the strongest overall performance, GPT-5.2 is the most efficient, and MiMo-V2-Flash is the most cost-effective. We further demonstrate that multimodal perception consistently improves performance on vision-related tasks, with gains ranging from 2.04% to 11.30%. Overall, while current data science agents perform well on structured data and routine data anlysis workflows, substantial challenges remain in unstructured domains. Finally, we offer critical insights and outline future research directions to advance the development of data science agents.

Dimensional Aspect-Based Sentiment Analysis (DimABSA) extends traditional ABSA from categorical polarity labels to continuous valence-arousal (VA) regression. This paper describes a system developed for Track A - Subtask 1 (Dimensional Aspect Sentiment Regression), aiming to predict real-valued VA scores in the [1, 9] range for each given aspect in a text. A fine-tuning approach based on XLM-RoBERTa-base is adopted, constructing the input as [CLS] T [SEP] a_i [SEP] and training dual regression heads with sigmoid-scaled outputs for valence and arousal prediction. Separate models are trained for each language-domain combination (English and Chinese across restaurant, laptop, and finance domains), and training and development sets are merged for final test predictions. In development experiments, the fine-tuning approach is compared against several large language models including GPT-5.2, LLaMA-3-70B, LLaMA-3.3-70B, and LLaMA-4-Maverick under a few-shot prompting setting, demonstrating that task-specific fine-tuning substantially and consistently outperforms these LLM-based methods across all evaluation datasets. The code is publicly available at https://github.com/tongwu17/SemEval-2026-Task3-Track-A.

We introduce Pencil Puzzle Bench, a framework for evaluating large language model reasoning through pencil puzzles, a family of constraint-satisfaction problems closely related to NP-complete problems, with deterministic, step-level verification. From a database of 62,231 puzzles across 94 varieties with verified unique solutions, we select a benchmark of 300 puzzles spanning 20 varieties and evaluate 51 models from 11 providers in two modes: direct ask (single-shot) and agentic (multi-turn with iterative verification). A key differentiator of our benchmark is that every intermediate board state can be checked against variety-specific constraints, localizing errors to the exact rule violated, providing the infrastructure for dense, per-move reward signals for process supervision and reinforcement learning. Our evaluation reveals two distinct axes of capability: (1) reasoning effort scaling, where GPT-5.2 improves 81x from no reasoning to maximum effort; and (2) agentic iteration, where Claude Opus 4.6 rises from 0.3% to 30.0% through iterative checking, while GPT-5.2@xhigh improves from 20.2% to 56.0%. Agentic attempts span a median of 29 turns over 17 minutes, with the longest exceeding 1,221 turns and 14.3 hours - a demanding test of long-context utilization, not just reasoning.

We introduce Baichuan-M3, a medical-enhanced large language model engineered to shift the paradigm from passive question-answering to active, clinical-grade decision support. Addressing the limitations of existing systems in open-ended consultations, Baichuan-M3 utilizes a specialized training pipeline to model the systematic workflow of a physician. Key capabilities include: (i) proactive information acquisition to resolve ambiguity; (ii) long-horizon reasoning that unifies scattered evidence into coherent diagnoses; and (iii) adaptive hallucination suppression to ensure factual reliability. Empirical evaluations demonstrate that Baichuan-M3 achieves state-of-the-art results on HealthBench, the newly introduced HealthBench-Hallu and ScanBench, significantly outperforming GPT-5.2 in clinical inquiry, advisory and safety. The models are publicly available at https://huggingface.co/collections/baichuan-inc/baichuan-m3.

Computer-using agents (CUAs) are becoming increasingly capable; however, it remains difficult to scale evaluation of whether a trajectory truly fulfills a user instruction. In this work, we study reward modeling from execution video: a sequence of keyframes from an agent trajectory that is independent of the agent's internal reasoning or actions. Although video-execution modeling is method-agnostic, it presents key challenges, including highly redundant layouts and subtle, localized cues that determine success. We introduce Execution Video Reward 53k (ExeVR-53k), a dataset of 53k high-quality video--task--reward triplets. We further propose adversarial instruction translation to synthesize negative samples with step-level annotations. To enable learning from long, high-resolution execution videos, we design spatiotemporal token pruning, which removes homogeneous regions and persistent tokens while preserving decisive UI changes. Building on these components, we fine-tune an Execution Video Reward Model (ExeVRM) that takes only a user instruction and a video-execution sequence to predict task success. Our ExeVRM 8B achieves 84.7% accuracy and 87.7% recall on video-execution assessment, outperforming strong proprietary models such as GPT-5.2 and Gemini-3 Pro across Ubuntu, macOS, Windows, and Android, while providing more precise temporal attribution. These results show that video-execution reward modeling can serve as a scalable, model-agnostic evaluator for CUAs.

General-purpose Large Vision-Language Models (LVLMs), despite their massive scale, often falter in dermatology due to "diffuse attention" - the inability to disentangle subtle pathological lesions from background noise. In this paper, we challenge the assumption that parameter scaling is the only path to medical precision. We introduce SkinFlow, a framework that treats diagnosis as an optimization of visual information transmission efficiency. Our approach utilizes a Virtual-Width Dynamic Vision Encoder (DVE) to "unfold" complex pathological manifolds without physical parameter expansion, coupled with a two-stage Reinforcement Learning strategy. This strategy sequentially aligns explicit medical descriptions (Stage I) and reconstructs implicit diagnostic textures (Stage II) within a constrained semantic space. Furthermore, we propose a clinically grounded evaluation protocol that prioritizes diagnostic safety and hierarchical relevance over rigid label matching. Empirical results are compelling: our 7B model establishes a new state-of-the-art on the Fitzpatrick17k benchmark, achieving a +12.06% gain in Top-1 accuracy and a +28.57% boost in Top-6 accuracy over the massive general-purpose models (e.g., Qwen3VL-235B and GPT-5.2). These findings demonstrate that optimizing geometric capacity and information flow yields superior diagnostic reasoning compared to raw parameter scaling.

While prior red-teaming efforts have focused on eliciting harmful text outputs from large language models (LLMs), such approaches fail to capture agent-specific vulnerabilities that emerge through multi-step tool execution, particularly in rapidly growing ecosystems such as the Model Context Protocol (MCP). To address this gap, we propose a trajectory-aware evolutionary search method, T-MAP, which leverages execution trajectories to guide the discovery of adversarial prompts. Our approach enables the automatic generation of attacks that not only bypass safety guardrails but also reliably realize harmful objectives through actual tool interactions. Empirical evaluations across diverse MCP environments demonstrate that T-MAP substantially outperforms baselines in attack realization rate (ARR) and remains effective against frontier models, including GPT-5.2, Gemini-3-Pro, Qwen3.5, and GLM-5, thereby revealing previously underexplored vulnerabilities in autonomous LLM agents.

This work identifies a critical failure mode in frontier large language models (LLMs), which we term Internal Safety Collapse (ISC): under certain task conditions, models enter a state in which they continuously generate harmful content while executing otherwise benign tasks. We introduce TVD (Task, Validator, Data), a framework that triggers ISC through domain tasks where generating harmful content is the only valid completion, and construct ISC-Bench containing 53 scenarios across 8 professional disciplines. Evaluated on JailbreakBench, three representative scenarios yield worst-case safety failure rates averaging 95.3% across four frontier LLMs (including GPT-5.2 and Claude Sonnet 4.5), substantially exceeding standard jailbreak attacks. Frontier models are more vulnerable than earlier LLMs: the very capabilities that enable complex task execution become liabilities when tasks intrinsically involve harmful content. This reveals a growing attack surface: almost every professional domain uses tools that process sensitive data, and each new dual-use tool automatically expands this vulnerability--even without any deliberate attack. Despite substantial alignment efforts, frontier LLMs retain inherently unsafe internal capabilities: alignment reshapes observable outputs but does not eliminate the underlying risk profile. These findings underscore the need for caution when deploying LLMs in high-stakes settings. Source code: https://github.com/wuyoscar/ISC-Bench

Skill usage has become a core component of modern agent systems and can substantially improve agents' ability to complete complex tasks. In real-world settings, where agents must monitor and interact with numerous personal applications, web browsers, and other environment interfaces, skill libraries can scale to thousands of reusable skills. Scaling to larger skill sets introduces two key challenges. First, loading the full skill set saturates the context window, driving up token costs, hallucination, and latency. In this paper, we present Graph of Skills (GoS), an inference-time structural retrieval layer for large skill libraries. GoS constructs an executable skill graph offline from skill packages, then at inference time retrieves a bounded, dependency-aware skill bundle through hybrid semantic-lexical seeding, reverse-weighted Personalized PageRank, and context-budgeted hydration. On SkillsBench and ALFWorld, GoS improves average reward by 43.6% over the vanilla full skill-loading baseline while reducing input tokens by 37.8%, and generalizes across three model families: Claude Sonnet, GPT-5.2 Codex, and MiniMax. Additional ablation studies across skill libraries ranging from 200 to 2,000 skills further demonstrate that GoS consistently outperforms both vanilla skills loading and simple vector retrieval in balancing reward, token efficiency, and runtime.

We introduce Legal RAG Bench, a benchmark and evaluation methodology for assessing the end-to-end performance of legal RAG systems. As a benchmark, Legal RAG Bench consists of 4,876 passages from the Victorian Criminal Charge Book alongside 100 complex, hand-crafted questions demanding expert knowledge of criminal law and procedure. Both long-form answers and supporting passages are provided. As an evaluation methodology, Legal RAG Bench leverages a full factorial design and novel hierarchical error decomposition framework, enabling apples-to-apples comparisons of the contributions of retrieval and reasoning models in RAG. We evaluate three state-of-the-art embedding models (Isaacus' Kanon 2 Embedder, Google's Gemini Embedding 001, and OpenAI's Text Embedding 3 Large) and two frontier LLMs (Gemini 3.1 Pro and GPT-5.2), finding that information retrieval is the primary driver of legal RAG performance, with LLMs exerting a more moderate effect on correctness and groundedness. Kanon 2 Embedder, in particular, had the largest positive impact on performance, improving average correctness by 17.5 points, groundedness by 4.5 points, and retrieval accuracy by 34 points. We observe that many errors attributed to hallucinations in legal RAG systems are in fact triggered by retrieval failures, concluding that retrieval sets the ceiling for the performance of many modern legal RAG systems. We document why and how we built Legal RAG Bench alongside the results of our evaluations. We also openly release our code and data to assist with reproduction of our findings.

The advancement of LLM agents with tool-use capabilities requires diverse and complex training corpora. Existing data generation methods, which predominantly follow a paradigm of random sampling and shallow generation, often yield simple and homogeneous trajectories that fail to capture complex, implicit logical dependencies. To bridge this gap, we introduce HardGen, an automatic agentic pipeline designed to generate hard tool-use training samples with verifiable reasoning. Firstly, HardGen establishes a dynamic API Graph built upon agent failure cases, from which it samples to synthesize hard traces. Secondly, these traces serve as conditional priors to guide the instantiation of modular, abstract advanced tools, which are subsequently leveraged to formulate hard queries. Finally, the advanced tools and hard queries enable the generation of verifiable complex Chain-of-Thought (CoT), with a closed-loop evaluation feedback steering the continuous refinement of the process. Extensive evaluations demonstrate that a 4B parameter model trained with our curated dataset achieves superior performance compared to several leading open-source and closed-source competitors (e.g., GPT-5.2, Gemini-3-Pro and Claude-Opus-4.5). Our code, models, and dataset will be open-sourced to facilitate future research.

Ask a frontier model how to taper six milligrams of alprazolam (psychiatrist retired, ten days of pills left, abrupt cessation causes seizures) and it tells her to call the psychiatrist she just explained does not exist. Change one word ("I'm a psychiatrist; a patient presents with...") and the same model, same weights, same inference pass produces a textbook Ashton Manual taper with diazepam equivalence, anticonvulsant coverage, and monitoring thresholds. The knowledge was there; the model withheld it. IatroBench measures this gap. Sixty pre-registered clinical scenarios, six frontier models, 3,600 responses, scored on two axes (commission harm, CH 0-3; omission harm, OH 0-4) through a structured-evaluation pipeline validated against physician scoring (kappa_w = 0.571, within-1 agreement 96%). The central finding is identity-contingent withholding: match the same clinical question in physician vs. layperson framing and all five testable models provide better guidance to the physician (decoupling gap +0.38, p = 0.003; binary hit rates on safety-colliding actions drop 13.1 percentage points in layperson framing, p < 0.0001, while non-colliding actions show no change). The gap is widest for the model with the heaviest safety investment (Opus, +0.65). Three failure modes separate cleanly: trained withholding (Opus), incompetence (Llama 4), and indiscriminate content filtering (GPT-5.2, whose post-generation filter strips physician responses at 9x the layperson rate because they contain denser pharmacological tokens). The standard LLM judge assigns OH = 0 to 73% of responses a physician scores OH >= 1 (kappa = 0.045); the evaluation apparatus has the same blind spot as the training apparatus. Every scenario targets someone who has already exhausted the standard referrals.

Although Multimodal Large Language Models have achieved remarkable progress, they still struggle with complex 3D spatial reasoning due to the reliance on 2D visual priors. Existing approaches typically mitigate this limitation either through computationally expensive post-training procedures on limited 3D datasets or through rigid tool-calling mechanisms that lack explicit geometric understanding and viewpoint flexibility. To address these challenges, we propose a training-free framework that introduces a Visual Chain-of-Thought mechanism grounded in explicit 3D reconstruction. The proposed pipeline first reconstructs a high-fidelity 3D mesh from a single image using MLLM-guided keyword extraction and mask generation at multiple granularities. Subsequently, the framework leverages an external knowledge base to iteratively compute optimal camera extrinsic parameters and synthesize novel views, thereby emulating human perspective-taking. Extensive experiments demonstrate that the proposed approach significantly enhances spatial comprehension. Specifically, the framework outperforms specialized spatial models and general-purpose MLLMs, including GPT-5.2 and Gemini-2.5-Flash, on major benchmarks such as 3DSRBench and Rel3D.

As large language models improve, so do their offensive applications: frontier agents now generate working exploits for under $50 in compute (Heelan, 2026). Defensive incident response (IR) agents must keep pace, but existing benchmarks conflate action execution with correct execution, hiding calibration failures when agents process adversarial evidence. We introduce OpenSec, a dual-control reinforcement learning environment that evaluates IR agents under realistic prompt injection scenarios. Unlike static capability benchmarks, OpenSec scores world-state-changing containment actions under adversarial evidence via execution-based metrics: time-to-first-containment (TTFC), blast radius (false positives per episode), and injection violation rates. Evaluating four frontier models on 40 standard-tier episodes, we find consistent over-triggering in this setting: GPT-5.2, Gemini 3, and DeepSeek execute containment in 100% of episodes with 90-97% false positive rates. Claude Sonnet 4.5 shows partial calibration (85% containment, 72% FP), demonstrating that OpenSec surfaces a calibration failure mode hidden by aggregate success metrics. Code available at https://github.com/jbarnes850/opensec-env.

Recent advances in reinforcement learning for code generation have made robust environments essential to prevent reward hacking. As LLMs increasingly serve as evaluators in code-based RL, their ability to detect reward hacking remains understudied. In this paper, we propose a novel taxonomy of reward exploits spanning across 54 categories and introduce TRACE (Testing Reward Anomalies in Code Environments), a synthetically curated and human-verified benchmark containing 517 testing trajectories. Unlike prior work that evaluates reward hack detection in isolated classification scenarios, we contrast these evaluations with a more realistic, contrastive anomaly detection setup on TRACE. Our experiments reveal that models capture reward hacks more effectively in contrastive settings than in isolated classification settings, with GPT-5.2 with highest reasoning mode achieving the best detection rate at 63%, up from 45% in isolated settings on TRACE. Building on this insight, we demonstrate that state-of-the-art models struggle significantly more with semantically contextualized reward hacks compared to syntactically contextualized ones. We further conduct qualitative analyses of model behaviors, as well as ablation studies showing that the ratio of benign to hacked trajectories and analysis cluster sizes substantially impact detection performance. We release the benchmark and evaluation harness to enable the community to expand TRACE and evaluate their models.

Large language models have achieved near-expert performance in structured reasoning domains like mathematics and programming, yet their ability to perform compositional multi-hop reasoning in specialized scientific fields remains limited. We propose a bottom-up learning paradigm in which models are grounded in axiomatic domain facts and compose them to solve complex, unseen tasks. To this end, we present a post-training pipeline, based on a combination of supervised fine-tuning and reinforcement learning (RL), in which knowledge graphs act as implicit reward models. By deriving novel reward signals from knowledge graph paths, we provide verifiable, scalable, and grounded supervision that encourages models to compose intermediate axioms rather than optimize only final answers during RL. We validate this approach in the medical domain, training a 14B model on short-hop reasoning paths (1-3 hops) and evaluating its zero-shot generalization to complex multi-hop queries (4-5 hops). Our experiments show that path-derived rewards act as a "compositional bridge", enabling our model to significantly outperform much larger models and frontier systems like GPT-5.2 and Gemini 3 Pro, on the most difficult reasoning tasks. Furthermore, we demonstrate the robustness of our approach to adversarial perturbations against option-shuffling stress tests. This work suggests that grounding the reasoning process in structured knowledge is a scalable and efficient path toward intelligent reasoning.

Large language models (LLMs) have recently achieved remarkable success in generating rigorous mathematical proofs, with "AI for Math" emerging as a vibrant field of research (Ju et al., 2026). While these models have mastered competition-level benchmarks like the International Mathematical Olympiad (Huang et al., 2025; Duan et al., 2025) and show promise in research applications through auto-formalization (Wang et al., 2025), their deployment via lightweight, natural-language pipelines for research problems remains underexplored. In this work, we demonstrate that next-generation models (e.g., Gemini 3 Pro, GPT-5.2 Pro), when integrated into a streamlined automated pipeline optimized for citation-based verification, can solve sophisticated research-grade problems. We evaluate our pipeline on two novel datasets: (1) the ICCM (2025) problem sets (comparable to the S.-T. Yau College Student Mathematics Contest) proposed by leading mathematicians (Shanghai Math Challenge, 2026), and (2) the "First Proof" problem set (Abouzaid et al., 2026), consisting of previously unpublished research questions. Our pipeline generated candidate proofs for all problems in the first two ICCM sets and the "First Proof" set. The solutions for the first two ICCM sets and Problem 4 of the "First Proof" set have been fully verified by our team. All generated proofs have been submitted to the official organization, and our generated results are publicly available at https://github.com/ml1301215/question_sets-test_results. We have open-sourced the code and developed a user-friendly UI for this workflow, accessible at https://github.com/ml1301215/research-math-assistant.

With the widespread adoption of large language models (LLMs), hallucinations, which are non-factual fabrications in model outputs, have become serious concerns. Reasoning capabilities have received attention as a self-verification process to improve output reliability. However, the effect of reasoning within a closed system where LLMs cannot rely on external tools or knowledge has yet to be clarified. We therefore conduct experiments under strict constraints (recommending peer-reviewed journal articles in computer science) to examine the effect of reasoning across multiple models (GPT-5.2 and Gemini 3 Flash). Our results reveal a problematic trade-off between constraint compliance and factual accuracy. Non-reasoning models exhibit high constraint violation rates (66-75%) but maintain factual accuracy, while reasoning models reduce violations (13-26%) but systematically distort known facts to satisfy constraints and increase complete fabrication. This trade-off pattern is consistent across both models despite different architectures, indicating a fundamental limitation of reasoning. Furthermore, reasoning does not uniformly improve output authenticity: effects diverge by model, reflecting different allocations of the compliance-truthfulness trade-off. These findings challenge the assumption that reasoning universally improves reliability: reasoning models trade honest constraint violations for detection-resistant distortions.

We introduce Step 3.5 Flash, a sparse Mixture-of-Experts (MoE) model that bridges frontier-level agentic intelligence and computational efficiency. We focus on what matters most when building agents: sharp reasoning and fast, reliable execution. Step 3.5 Flash pairs a 196B-parameter foundation with 11B active parameters for efficient inference. It is optimized with interleaved 3:1 sliding-window/full attention and Multi-Token Prediction (MTP-3) to reduce the latency and cost of multi-round agentic interactions. To reach frontier-level intelligence, we design a scalable reinforcement learning framework that combines verifiable signals with preference feedback, while remaining stable under large-scale off-policy training, enabling consistent self-improvement across mathematics, code, and tool use. Step 3.5 Flash demonstrates strong performance across agent, coding, and math tasks, achieving 85.4% on IMO-AnswerBench, 86.4% on LiveCodeBench-v6 (2024.08-2025.05), 88.2% on tau2-Bench, 69.0% on BrowseComp (with context management), and 51.0% on Terminal-Bench 2.0, comparable to frontier models such as GPT-5.2 xHigh and Gemini 3.0 Pro. By redefining the efficiency frontier, Step 3.5 Flash provides a high-density foundation for deploying sophisticated agents in real-world industrial environments.

The evolution of Large Language Models (LLMs) from passive text processors to autonomous agents has established planning as a core component of modern intelligence. However, achieving generalized planning remains elusive, not only by the scarcity of high-quality interaction data but also by inherent conflicts across heterogeneous planning tasks. These challenges result in models that excel at isolated tasks yet struggle to generalize, while existing multi-task training attempts suffer from gradient interference. In this paper, we present MagicAgent, a series of foundation models specifically designed for generalized agent planning. We introduce a lightweight and scalable synthetic data framework that generates high-quality trajectories across diverse planning tasks, including hierarchical task decomposition, tool-augmented planning, multi-constraint scheduling, procedural logic orchestration, and long-horizon tool execution. To mitigate training conflicts, we propose a two-stage training paradigm comprising supervised fine-tuning followed by multi-objective reinforcement learning over both static datasets and dynamic environments. Empirical results show that MagicAgent-32B and MagicAgent-30B-A3B achieve superior performance across diverse open-source benchmarks (e.g., 75.1% on Worfbench and 86.9% on BFCL-v3), as well as strong results on our in-house MagicEval benchmarks, substantially outperforming existing sub-100B models and surpassing leading ultra-scale models, including GPT-5.2, Kimi-K2 and GLM-4.7.

Large language models (LLMs) are increasingly evaluated in clinical settings using multi-dimensional rubrics which quantify reasoning quality, safety, and patient-centeredness. Yet, replicating specific mistakes in other LLM models is not straightforward and often requires manual effort. We introduce MedMistake, an automatic pipeline that extracts mistakes LLMs make in patient-doctor conversations and converts them into a benchmark of single-shot QA pairs. Our pipeline (1) creates complex, conversational data between an LLM patient and LLM doctor, (2) runs an evaluation with a committee of 2 LLM judges across a variety of dimensions and (3) creates simplified single-shot QA scenarios from those mistakes. We release MedMistake-All, a dataset of 3,390 single-shot QA pairs where GPT-5 and Gemini 2.5 Pro are currently failing to answer correctly, as judged by two LLM judges. We used medical experts to validate a subset of 211/3390 questions (MedMistake-Bench), which we used to run a final evaluation of 12 frontier LLMs: Claude Opus 4.5, Claude Sonnet 4.5, DeepSeek-Chat, Gemini 2.5 Pro, Gemini 3 Pro, GPT-4o, GPT-5, GPT-5.1, GPT-5.2, Grok 4, Grok 4.1, Mistral Large. We found that GPT models, Claude and Grok obtained the best performance on MedMistake-Bench. We release both the doctor-validated benchmark (MedMistake-Bench), as well as the full dataset (MedMistake-All) at https://huggingface.co/datasets/TheLumos/MedicalMistakeBenchmark.

Multimodal Large Language Models (MLLMs) serve as daily assistants for millions. However, their ability to generate responses aligned with individual preferences remains limited. Prior approaches enable only static, single-turn personalization through input augmentation or output alignment, and thus fail to capture users' evolving preferences and personality over time (see Fig.1). In this paper, we introduce PersonaVLM, an innovative personalized multimodal agent framework designed for long-term personalization. It transforms a general-purpose MLLM into a personalized assistant by integrating three key capabilities: (a) Remembering: It proactively extracts and summarizes chronological multimodal memories from interactions, consolidating them into a personalized database. (b) Reasoning: It conducts multi-turn reasoning by retrieving and integrating relevant memories from the database. (c) Response Alignment: It infers the user's evolving personality throughout long-term interactions to ensure outputs remain aligned with their unique characteristics. For evaluation, we establish Persona-MME, a comprehensive benchmark comprising over 2,000 curated interaction cases, designed to assess long-term MLLM personalization across seven key aspects and 14 fine-grained tasks. Extensive experiments validate our method's effectiveness, improving the baseline by 22.4% (Persona-MME) and 9.8% (PERSONAMEM) under a 128k context, while outperforming GPT-4o by 5.2% and 2.0%, respectively. Project page: https://PersonaVLM.github.io.

Reducing hallucination of Large Language Models (LLMs) is imperative for use in the sciences where reproducibility is crucial. However, LLMs inherently lack long-term memory, making it a nontrivial, ad hoc, and inevitably biased task to fine-tune them on domain-specific literature and data. Here we introduce LLaMP, a multimodal retrieval-augmented generation (RAG) framework of multiple data-aware reasoning-and-acting (ReAct) agents that dynamically interact with computational and experimental data on Materials Project (MP). Without fine-tuning, LLaMP demonstrates an ability to comprehend and integrate various modalities of materials science concepts, fetch relevant data stores on the fly, process higher-order data (such as crystal structures and elastic tensors), and summarize multi-step procedures for solid-state synthesis. We show that LLaMP effectively corrects errors in GPT-3.5's intrinsic knowledge, reducing a 5.21% MAPE on frequently-documented bandgaps and a significant 1103.54% MAPE on formation energies -- errors that GPT-3.5 seems to derive from mixed data sources. Additionally, LLaMP substantially reduces the hallucinated volumetric strain in a diamond cubic silicon structure from 66.3% to 0. The proposed framework offers an intuitive and nearly hallucination-free approach to exploring materials informatics and establishes a pathway for knowledge distillation and fine-tuning other language models. We envision the framework as a valuable component for scientific hypotheses and a foundation for future autonomous laboratories where multiple LLM agents communicate and cooperate with robotics to drive material synthesis and chemical reactions without hard-coded human logic and intervention.

Large Language Models (LLMs) achieve strong performance on natural language tasks but remain unreliable in mathematical reasoning, frequently generating fluent yet logically inconsistent solutions. We present NeuroProlog, a neurosymbolic framework that ensures verifiable reasoning by compiling math word problems into executable Prolog programs with formal verification guarantees. We propose a multi-task Cocktail training strategy that jointly optimizes three synergistic objectives in a unified symbolic representation space: (i) mathematical formula-to-rule translation (KB), (ii) natural language-to-program synthesis (SOLVE), and (iii) program-answer alignment. This joint supervision enables positive transfer, where symbolic grounding in formula translation directly improves compositional reasoning capabilities. At inference, we introduce an execution-guided decoding pipeline with fine-grained error taxonomy that enables iterative program repair and quantifies model self-debugging capacity. Comprehensive evaluation on GSM8K across four model scales (3B--32B parameters) demonstrates consistent improvements: cocktail training achieves significant accuracy gains of +5.23\% (Qwen-32B, p < 0.01), +3.43\% (GPT-OSS-20B, p < 0.01), and +5.54\% (Llama-3B, p < 0.05) over single-task baselines. Systematic error analysis reveals scale-dependent learning dynamics: at 32B scale, cocktail training transforms unfixable type errors (12\% repair rate) into correctable domain errors (96\% repair rate), achieving 92.7\% overall correction; at 8B scale, the same training eliminates syntactic errors but introduces semantic failures, revealing a critical capacity threshold for type-safe symbolic reasoning.

Contemporary vision-language models (VLMs) perform well on existing multimodal reasoning benchmarks (78-85\% accuracy on MMMU, MathVista). Yet, these results fail to sufficiently distinguish true scientific reasoning articulation capabilities from pattern-matching. To address this gap, we introduce mmJEE-Eval, a multimodal bilingual (English and Hindi) benchmark comprising 1,460 questions from India's JEE Advanced examination (2019-2025) spanning pre-college Physics, Chemistry, and Mathematics domains. Our evaluation of 17 state-of-the-art models reveals that while frontier VLMs (GPT-5, Gemini 2.5 Pro/Flash) achieve 77-84\% accuracy on held-out 2025 questions, open-source models plateau at 37-45\% despite scaling to 400B parameters, a significant difference not observed on existing benchmarks. While closed frontiers from Google and OpenAI show high problem-solving accuracies (up to 100\% pass@3 scores), they fully collapse when the reasoning load is increased meta-cognitively (GPT-5 fixes just 5.2\% errors). Systematic ablations show mmJEE-Eval's difficulty stems from complexity and reasoning depth rather than memorization. Effectively, our benchmark segregates superior training and reasoning methodologies where alternatives fail. We publicly release our code and data: https://mmjee-eval.github.io