Daily Papers

Recent work on chain-of-thought (CoT) faithfulness reports single aggregate numbers (e.g., DeepSeek-R1 acknowledges hints 39% of the time), implying that faithfulness is an objective, measurable property of a model. This paper demonstrates that it is not. Three classifiers (a regex-only detector, a two-stage regex-plus-LLM pipeline, and an independent Claude Sonnet 4 judge) are applied to 10,276 influenced reasoning traces from 12 open-weight models spanning 9 families and 7B to 1T parameters. On identical data, these classifiers produce overall faithfulness rates of 74.4%, 82.6%, and 69.7%, respectively, with non-overlapping 95% confidence intervals. Per-model gaps range from 2.6 to 30.6 percentage points; all are statistically significant (McNemar's test, p < 0.001). The disagreements are systematic, not random: inter-classifier agreement measured by Cohen's kappa ranges from 0.06 ("slight") for sycophancy hints to 0.42 ("moderate") for grader hints, and the asymmetry is pronounced: for sycophancy, 883 cases are classified as faithful by the pipeline but unfaithful by the Sonnet judge, while only 2 go the other direction. Classifier choice can also reverse model rankings: Qwen3.5-27B ranks 1st under the pipeline but 7th under the Sonnet judge; OLMo-3.1-32B moves in the opposite direction, from 9th to 3rd. The root cause is that different classifiers operationalize related faithfulness constructs at different levels of stringency (lexical mention versus epistemic dependence), and these constructs yield divergent measurements on the same behavior. These results demonstrate that published faithfulness numbers cannot be meaningfully compared across studies that use different classifiers, and that future evaluations should report sensitivity ranges across multiple classification methodologies rather than single point estimates.

Seeking answers to questions within long scientific research articles is a crucial area of study that aids readers in quickly addressing their inquiries. However, existing question-answering (QA) datasets based on scientific papers are limited in scale and focus solely on textual content. To address this limitation, we introduce SPIQA (Scientific Paper Image Question Answering), the first large-scale QA dataset specifically designed to interpret complex figures and tables within the context of scientific research articles across various domains of computer science. Leveraging the breadth of expertise and ability of multimodal large language models (MLLMs) to understand figures, we employ automatic and manual curation to create the dataset. We craft an information-seeking task involving multiple images that cover a wide variety of plots, charts, tables, schematic diagrams, and result visualizations. SPIQA comprises 270K questions divided into training, validation, and three different evaluation splits. Through extensive experiments with 12 prominent foundational models, we evaluate the ability of current multimodal systems to comprehend the nuanced aspects of research articles. Additionally, we propose a Chain-of-Thought (CoT) evaluation strategy with in-context retrieval that allows fine-grained, step-by-step assessment and improves model performance. We further explore the upper bounds of performance enhancement with additional textual information, highlighting its promising potential for future research and the dataset's impact on revolutionizing how we interact with scientific literature.

Learning from negative samples holds great promise for improving Large Language Model (LLM) reasoning capability, yet existing methods treat all incorrect responses as equally informative, overlooking the crucial role of sample quality. To address this, we propose Plausible Negative Samples (PNS), a method that synthesizes high-quality negative samples exhibiting expected format and structural coherence while ultimately yielding incorrect answers. PNS trains a dedicated model via reverse reinforcement learning (RL) guided by a composite reward combining format compliance, accuracy inversion, reward model assessment, and chain-of-thought evaluation, generating responses nearly indistinguishable from correct solutions. We further validate PNS as a plug-and-play data source for preference optimization across three backbone models on seven mathematical reasoning benchmarks. Results demonstrate that PNS consistently outperforms other negative sample synthesis methods, achieving an average improvement of 2.03% over RL-trained models.

The remarkable progress of Multi-modal Large Language Models (MLLMs) has garnered unparalleled attention, due to their superior performance in visual contexts. However, their capabilities in visual math problem-solving remain insufficiently evaluated and understood. We investigate current benchmarks to incorporate excessive visual content within textual questions, which potentially assist MLLMs in deducing answers without truly interpreting the input diagrams. To this end, we introduce MathVerse, an all-around visual math benchmark designed for an equitable and in-depth evaluation of MLLMs. We meticulously collect 2,612 high-quality, multi-subject math problems with diagrams from publicly available sources. Each problem is then transformed by human annotators into six distinct versions, each offering varying degrees of information content in multi-modality, contributing to 15K test samples in total. This approach allows MathVerse to comprehensively assess whether and how much MLLMs can truly understand the visual diagrams for mathematical reasoning. In addition, we propose a Chain-of-Thought (CoT) evaluation strategy for a fine-grained assessment of the output answers. Rather than naively judging True or False, we employ GPT-4(V) to adaptively extract crucial reasoning steps, and then score each step with detailed error analysis, which can reveal the intermediate CoT reasoning quality by MLLMs. We hope the MathVerse benchmark may provide unique insights to guide the future development of MLLMs. Project page: https://mathverse-cuhk.github.io

The recent advancements in large language models (LLMs) have revolutionized the field of natural language processing, progressively broadening their scope to multimodal perception and generation. However, effectively integrating listening capabilities into LLMs poses significant challenges, particularly with respect to generalizing across varied contexts and executing complex auditory tasks. In this work, we introduce WavLLM, a robust and adaptive speech large language model with dual encoders, and a prompt-aware LoRA weight adapter, optimized by a two-stage curriculum learning approach. Leveraging dual encoders, we decouple different types of speech information, utilizing a Whisper encoder to process the semantic content of speech, and a WavLM encoder to capture the unique characteristics of the speaker's identity. Within the curriculum learning framework, WavLLM first builds its foundational capabilities by optimizing on mixed elementary single tasks, followed by advanced multi-task training on more complex tasks such as combinations of the elementary tasks. To enhance the flexibility and adherence to different tasks and instructions, a prompt-aware LoRA weight adapter is introduced in the second advanced multi-task training stage. We validate the proposed model on universal speech benchmarks including tasks such as ASR, ST, SV, ER, and also apply it to specialized datasets like Gaokao English listening comprehension set for SQA, and speech Chain-of-Thought (CoT) evaluation set. Experiments demonstrate that the proposed model achieves state-of-the-art performance across a range of speech tasks on the same model size, exhibiting robust generalization capabilities in executing complex tasks using CoT approach. Furthermore, our model successfully completes Gaokao tasks without specialized training. The codes, models, audio, and Gaokao evaluation set can be accessed at aka.ms/wavllm.

Large Language Models (LLMs) often exhibit hallucinations, generating factually incorrect or semantically irrelevant content in response to prompts. Chain-of-Thought (CoT) prompting can mitigate hallucinations by encouraging step-by-step reasoning, but its impact on hallucination detection remains underexplored. To bridge this gap, we conduct a systematic empirical evaluation. We begin with a pilot experiment, revealing that CoT reasoning significantly affects the LLM's internal states and token probability distributions. Building on this, we evaluate the impact of various CoT prompting methods on mainstream hallucination detection methods across both instruction-tuned and reasoning-oriented LLMs. Specifically, we examine three key dimensions: changes in hallucination score distributions, variations in detection accuracy, and shifts in detection confidence. Our findings show that while CoT prompting helps reduce hallucination frequency, it also tends to obscure critical signals used for detection, impairing the effectiveness of various detection methods. Our study highlights an overlooked trade-off in the use of reasoning. Code is publicly available at: https://github.com/ECNU-Text-Computing/cot-hallu-detect .

The advancement of Chain-of-Thought (CoT) reasoning has significantly enhanced the capabilities of large language models (LLMs) and large vision-language models (LVLMs). However, a rigorous evaluation framework for video CoT reasoning remains absent. Current video benchmarks fail to adequately assess the reasoning process and expose whether failures stem from deficiencies in perception or reasoning capabilities. Therefore, we introduce VCR-Bench, a novel benchmark designed to comprehensively evaluate LVLMs' Video Chain-of-Thought Reasoning capabilities. VCR-Bench comprises 859 videos spanning a variety of video content and durations, along with 1,034 high-quality question-answer pairs. Each pair is manually annotated with a stepwise CoT rationale, where every step is tagged to indicate its association with the perception or reasoning capabilities. Furthermore, we design seven distinct task dimensions and propose the CoT score to assess the entire CoT process based on the stepwise tagged CoT rationals. Extensive experiments on VCR-Bench highlight substantial limitations in current LVLMs. Even the top-performing model, o1, only achieves a 62.8% CoT score and an 56.7% accuracy, while most models score below 40%. Experiments show most models score lower on perception than reasoning steps, revealing LVLMs' key bottleneck in temporal-spatial information processing for complex video reasoning. A robust positive correlation between the CoT score and accuracy confirms the validity of our evaluation framework and underscores the critical role of CoT reasoning in solving complex video reasoning tasks. We hope VCR-Bench to serve as a standardized evaluation framework and expose the actual drawbacks in complex video reasoning task.

Automatic mean opinion score (MOS) prediction provides a more perceptual alternative to objective metrics, offering deeper insights into the evaluated models. With the rapid progress of multimodal large language models (MLLMs), their enhanced perceptual and reasoning abilities enable more comprehensive and interpretable audio quality assessment. In this work, we tackle the challenging task of audio editing evaluation and propose the first natural language-based automated evaluation framework built on MLLMs. Our approach introduces two fine-tuning tasks to boost multi-audio understanding, combined with Chain-of-Thought prompting, and lightweight instruction tuning, to enhance step-by-step reasoning. Experiment demonstrate that our framework delivers accurate, interpretable, and text-based editing evaluation, closely aligning with human judgments and objective metrics while substantially improving over baselines. The code and demo are available at https://github.com/NKU-HLT/Eval_Reasoning.

Large language models (LLMs) are increasingly used as judges to evaluate agent performance, particularly in non-verifiable settings where judgments rely on agent trajectories including chain-of-thought (CoT) reasoning. This paradigm implicitly assumes that the agent's CoT faithfully reflects both its internal reasoning and the underlying environment state. We show this assumption is brittle: LLM judges are highly susceptible to manipulation of agent reasoning traces. By systematically rewriting agent CoTs while holding actions and observations fixed, we demonstrate that manipulated reasoning alone can inflate false positive rates of state-of-the-art VLM judges by up to 90% across 800 trajectories spanning diverse web tasks. We study manipulation strategies spanning style-based approaches that alter only the presentation of reasoning and content-based approaches that fabricate signals of task progress, and find that content-based manipulations are consistently more effective. We evaluate prompting-based techniques and scaling judge-time compute, which reduce but do not fully eliminate susceptibility to manipulation. Our findings reveal a fundamental vulnerability in LLM-based evaluation and highlight the need for judging mechanisms that verify reasoning claims against observable evidence.

Multimodal Reasoning Models (MRMs) leveraging Chain-of-Thought (CoT) based thinking have revolutionized mathematical and logical problem-solving. However, we show that this paradigm struggles with generalized spatial intelligence. We perform a comprehensive evaluation of seventeen models across thirteen spatial benchmarks and identify a critical gap: CoT prompting consistently degrades performance in visual spatial reasoning. Furthermore, through a novel No-Image++ ablation, we demonstrate that MRMs and CoT prompted MLMs suffer from severe shortcut learning, and hallucinate visual details from textual priors even when the image is absent. These findings challenge the efficacy of text-only CoT for spatial tasks and underscore the need for vision-centric reasoning paradigms.

Chain-of-thought (CoT) outputs let us read a model's step-by-step reasoning. Since any long, serial reasoning process must pass through this textual trace, the quality of the CoT is a direct window into what the model is thinking. This visibility could help us spot unsafe or misaligned behavior (monitorability), but only if the CoT is transparent about its internal reasoning (faithfulness). Fully measuring faithfulness is difficult, so researchers often focus on examining the CoT in cases where the model changes its answer after adding a cue to the input. This proxy finds some instances of unfaithfulness but loses information when the model maintains its answer, and does not investigate aspects of reasoning not tied to the cue. We extend these results to a more holistic sense of monitorability by introducing verbosity: whether the CoT lists every factor needed to solve the task. We combine faithfulness and verbosity into a single monitorability score that shows how well the CoT serves as the model's external `working memory', a property that many safety schemes based on CoT monitoring depend on. We evaluate instruction-tuned and reasoning models on BBH, GPQA, and MMLU. Our results show that models can appear faithful yet remain hard to monitor when they leave out key factors, and that monitorability differs sharply across model families. We release our evaluation code using the Inspect library to support reproducible future work.

Image retrieval remains a fundamental yet challenging problem in computer vision. While recent advances in Multimodal Large Language Models (MLLMs) have demonstrated strong reasoning capabilities, existing methods typically employ them only for evaluation, without involving them directly in the ranking process. As a result, their rich multimodal reasoning abilities remain underutilized, leading to suboptimal performance. In this paper, we propose a novel Chain-of-Thought Re-Ranking (CoTRR) method to address this issue. Specifically, we design a listwise ranking prompt that enables MLLM to directly participate in re-ranking candidate images. This ranking process is grounded in an image evaluation prompt, which assesses how well each candidate aligns with users query. By allowing MLLM to perform listwise reasoning, our method supports global comparison, consistent reasoning, and interpretable decision-making - all of which are essential for accurate image retrieval. To enable structured and fine-grained analysis, we further introduce a query deconstruction prompt, which breaks down the original query into multiple semantic components. Extensive experiments on five datasets demonstrate the effectiveness of our CoTRR method, which achieves state-of-the-art performance across three image retrieval tasks, including text-to-image retrieval (TIR), composed image retrieval (CIR) and chat-based image retrieval (Chat-IR). Our code is available at https://github.com/freshfish15/CoTRR .

Large Language Models (LLMs) (e.g., ChatGPT) have shown impressive performance in code generation. LLMs take prompts as inputs, and Chain-of-Thought (CoT) prompting is the state-of-the-art prompting technique. CoT prompting asks LLMs first to generate CoTs (i.e., intermediate natural language reasoning steps) and then output the code. However, CoT prompting is designed for natural language generation and has low accuracy in code generation. In this paper, we propose Structured CoTs (SCoTs) and present a novel prompting technique for code generation, named SCoT prompting. Our motivation is source code contains rich structural information and any code can be composed of three program structures (i.e., sequence, branch, and loop structures). Intuitively, structured intermediate reasoning steps make for structured source code. Thus, we ask LLMs to use program structures to build CoTs, obtaining SCoTs. Then, LLMs generate the final code based on SCoTs. Compared to CoT prompting, SCoT prompting explicitly constrains LLMs to think about how to solve requirements from the view of source code and further the performance of LLMs in code generation. We apply SCoT prompting to two LLMs (i.e., ChatGPT and Codex) and evaluate it on three benchmarks (i.e., HumanEval, MBPP, and MBCPP). (1) SCoT prompting outperforms the state-of-the-art baseline - CoT prompting by up to 13.79% in Pass@1. (2) Human evaluation shows human developers prefer programs from SCoT prompting. (3) SCoT prompting is robust to examples and achieves substantial improvements.

Prompting language models to provide step-by-step answers (e.g., "Chain-of-Thought") is the prominent approach for complex reasoning tasks, where more accurate reasoning chains typically improve downstream task performance. Recent literature discusses automatic methods to verify reasoning steps to evaluate and improve their correctness. However, no fine-grained step-level datasets are available to enable thorough evaluation of such verification methods, hindering progress in this direction. We introduce Reveal: Reasoning Verification Evaluation, a new dataset to benchmark automatic verifiers of complex Chain-of-Thought reasoning in open-domain question answering settings. Reveal includes comprehensive labels for the relevance, attribution to evidence passages, and logical correctness of each reasoning step in a language model's answer, across a wide variety of datasets and state-of-the-art language models.

Despite great progress, existing multimodal large language models (MLLMs) are prone to visual hallucination, greatly impeding their trustworthy applications. In this paper, we study this problem from the perspective of visual-spatial reasoning, and propose a new learning task for MLLMs, termed Grounded Chain-of-Thought (GCoT). Different from recent visual CoT studies, which focus more on visual knowledge reasoning, GCoT is keen to helping MLLMs to recognize and ground the relevant visual cues step by step, thereby predicting the correct answer with grounding coordinates as the intuitive basis. To facilitate this task, we also carefully design and construct a dataset called multimodal grounded chain-of-thought (MM-GCoT) consisting of 24,022 GCoT examples for 5,033 images. Besides, a comprehensive consistency evaluation system is also introduced, including the metrics of answer accuracy, grounding accuracy and answer-grounding consistency. We further design and conduct a bunch of experiments on 12 advanced MLLMs, and reveal some notable findings: i. most MLLMs performs poorly on the consistency evaluation, indicating obvious visual hallucination; ii. visual hallucination is not directly related to the parameter size and general multimodal performance, i.e., a larger and stronger MLLM is not less affected by this issue. Lastly, we also demonstrate that the proposed dataset can help existing MLLMs to well cultivate their GCoT capability and reduce the inconsistent answering significantly. Moreover, their GCoT can be also generalized to exiting multimodal tasks, such as open-world QA and REC.

Chain-of-Thought (CoT) reasoning has proven effective in natural language tasks but remains underexplored in multimodal alignment. This study investigates its integration into 3D vision-language learning by embedding structured reasoning into alignment training. We introduce the 3D-CoT Benchmark, a dataset with hierarchical CoT annotations covering shape recognition, functional inference, and causal reasoning. Through controlled experiments, we compare CoT-structured and standard textual annotations across large reasoning models (LRMs) and large language models (LLMs). Our evaluation employs a dual-layer framework assessing both intermediate reasoning and final inference quality. Extensive experiments demonstrate that CoT significantly improves 3D semantic grounding, with LRMs leveraging CoT more effectively than LLMs. Furthermore, we highlight that annotation structure influences performance-explicit reasoning markers aid LLMs, while unmarked CoT better aligns with LRM inference patterns. Our analyses suggest that CoT is crucial for enhancing multimodal reasoning, with implications beyond 3D tasks.

State-of-the-art performance in QA tasks is currently achieved by systems employing Large Language Models (LLMs), however these models tend to hallucinate information in their responses. One approach focuses on enhancing the generation process by incorporating attribution from the given input to the output. However, the challenge of identifying appropriate attributions and verifying their accuracy against a source is a complex task that requires significant improvements in assessing such systems. We introduce an attribution-oriented Chain-of-Thought reasoning method to enhance the accuracy of attributions. This approach focuses the reasoning process on generating an attribution-centric output. Evaluations on two context-enhanced question-answering datasets using GPT-4 demonstrate improved accuracy and correctness of attributions. In addition, the combination of our method with finetuning enhances the response and attribution accuracy of two smaller LLMs, showing their potential to outperform GPT-4 in some cases.

While Large Language Models (LLMs) have excelled in textual reasoning, they struggle with mathematical domains like geometry that intrinsically rely on visual aids. Existing approaches to Visual Chain-of-Thought (VCoT) are often limited by rigid external tools or fail to generate the high-fidelity, strategically-timed diagrams necessary for complex problem-solving. To bridge this gap, we introduce MathCanvas, a comprehensive framework designed to endow unified Large Multimodal Models (LMMs) with intrinsic VCoT capabilities for mathematics. Our approach consists of two phases. First, a Visual Manipulation stage pre-trains the model on a novel 15.2M-pair corpus, comprising 10M caption-to-diagram pairs (MathCanvas-Imagen) and 5.2M step-by-step editing trajectories (MathCanvas-Edit), to master diagram generation and editing. Second, a Strategic Visual-Aided Reasoning stage fine-tunes the model on MathCanvas-Instruct, a new 219K-example dataset of interleaved visual-textual reasoning paths, teaching it when and how to leverage visual aids. To facilitate rigorous evaluation, we introduce MathCanvas-Bench, a challenging benchmark with 3K problems that require models to produce interleaved visual-textual solutions. Our model, BAGEL-Canvas, trained under this framework, achieves an 86% relative improvement over strong LMM baselines on MathCanvas-Bench, demonstrating excellent generalization to other public math benchmarks. Our work provides a complete toolkit-framework, datasets, and benchmark-to unlock complex, human-like visual-aided reasoning in LMMs. Project Page: https://mathcanvas.github.io/

Chain-of-thought prompting combined with pre-trained large language models has achieved encouraging results on complex reasoning tasks. In this paper, we propose a new decoding strategy, self-consistency, to replace the naive greedy decoding used in chain-of-thought prompting. It first samples a diverse set of reasoning paths instead of only taking the greedy one, and then selects the most consistent answer by marginalizing out the sampled reasoning paths. Self-consistency leverages the intuition that a complex reasoning problem typically admits multiple different ways of thinking leading to its unique correct answer. Our extensive empirical evaluation shows that self-consistency boosts the performance of chain-of-thought prompting with a striking margin on a range of popular arithmetic and commonsense reasoning benchmarks, including GSM8K (+17.9%), SVAMP (+11.0%), AQuA (+12.2%), StrategyQA (+6.4%) and ARC-challenge (+3.9%).

Large Language Models (LLMs) have shown impressive performance in complex reasoning tasks through Chain-of-Thought (CoT) reasoning, allowing models to break down problems into manageable sub-tasks. However, existing CoT evaluation techniques either require annotated CoT data or fall short in accurately assessing intermediate reasoning steps, leading to high rates of false positives. In this paper, we formalize CoT reasoning in LLMs through an information-theoretic lens. Specifically, our framework quantifies the `information gain' at each reasoning step, enabling the identification of failure modes in LLMs without the need for expensive annotated datasets. We demonstrate the efficacy of our approach through extensive experiments on toy and GSM-8K data, where it significantly outperforms existing outcome-based methods by providing more accurate insights into model performance on individual tasks.

As large language models (LLMs) are continuously being developed, their evaluation becomes increasingly important yet challenging. This work proposes Chain-of-Thought Hub, an open-source evaluation suite on the multi-step reasoning capabilities of large language models. We are interested in this setting for two reasons: (1) from the behavior of GPT and PaLM model family, we observe that complex reasoning is likely to be a key differentiator between weaker and stronger LLMs; (2) we envisage large language models to become the next-generation computational platform and foster an ecosystem of LLM-based new applications, this naturally requires the foundation models to perform complex tasks that often involve the composition of linguistic and logical operations. Our approach is to compile a suite of challenging reasoning benchmarks to track the progress of LLMs. Our current results show that: (1) model scale clearly correlates with reasoning capabilities; (2) As of May 2023, Claude-v1.3 and PaLM-2 are the only two models that are comparable with GPT-4, while open-sourced models still lag behind; (3) LLaMA-65B performs closely to code-davinci-002, indicating that with successful further development such as reinforcement learning from human feedback (RLHF), it has great potential to be close to GPT-3.5-Turbo. Our results also suggest that for the open-source efforts to catch up, the community may focus more on building better base models and exploring RLHF.

Identifying procedural errors online from egocentric videos is a critical yet challenging task across various domains, including manufacturing, healthcare, and skill-based training. The nature of such mistakes is inherently open-set, as unforeseen or novel errors may occur, necessitating robust detection systems that do not rely on prior examples of failure. Currently, however, no technique effectively detects open-set procedural mistakes online. We propose a dual branch architecture to address this problem in an online fashion: one branch continuously performs step recognition from the input egocentric video, while the other anticipates future steps based on the recognition module's output. Mistakes are detected as mismatches between the currently recognized action and the action predicted by the anticipation module. The recognition branch takes input frames, predicts the current action, and aggregates frame-level results into action tokens. The anticipation branch, specifically, leverages the solid pattern-matching capabilities of Large Language Models (LLMs) to predict action tokens based on previously predicted ones. Given the online nature of the task, we also thoroughly benchmark the difficulties associated with per-frame evaluations, particularly the need for accurate and timely predictions in dynamic online scenarios. Extensive experiments on two procedural datasets demonstrate the challenges and opportunities of leveraging a dual-branch architecture for mistake detection, showcasing the effectiveness of our proposed approach. In a thorough evaluation including recognition and anticipation variants and state-of-the-art models, our method reveals its robustness and effectiveness in online applications.

Recent work, using the Biasing Features metric, labels a CoT as unfaithful if it omits a prompt-injected hint that affected the prediction. We argue this metric confuses unfaithfulness with incompleteness, the lossy compression needed to turn distributed transformer computation into a linear natural language narrative. On multi-hop reasoning tasks with Llama-3 and Gemma-3, many CoTs flagged as unfaithful by Biasing Features are judged faithful by other metrics, exceeding 50% in some models. With a new faithful@k metric, we show that larger inference-time token budgets greatly increase hint verbalization (up to 90% in some settings), suggesting much apparent unfaithfulness is due to tight token limits. Using Causal Mediation Analysis, we further show that even non-verbalized hints can causally mediate prediction changes through the CoT. We therefore caution against relying solely on hint-based evaluations and advocate a broader interpretability toolkit, including causal mediation and corruption-based metrics.

While chain-of-thought (CoT) reasoning improves the performance of large language models (LLMs) in complex tasks, it still has two main challenges: the low reliability of relying solely on LLMs to generate reasoning chains and the interference of natural language reasoning chains on the inference logic of LLMs. To address these issues, we propose CoT-RAG, a novel reasoning framework with three key designs: (i) Knowledge Graph-driven CoT Generation, featuring knowledge graphs to modulate reasoning chain generation of LLMs, thereby enhancing reasoning credibility; (ii) Learnable Knowledge Case-aware RAG, which incorporates retrieval-augmented generation (RAG) into knowledge graphs to retrieve relevant sub-cases and sub-descriptions, providing LLMs with learnable information; (iii) Pseudo-Program Prompting Execution, which encourages LLMs to execute reasoning tasks in pseudo-programs with greater logical rigor. We conduct a comprehensive evaluation on nine public datasets, covering three reasoning problems. Compared with the-state-of-the-art methods, CoT-RAG exhibits a significant accuracy improvement, ranging from 4.0% to 23.0%. Furthermore, testing on four domain-specific datasets, CoT-RAG shows remarkable accuracy and efficient execution, highlighting its strong practical applicability and scalability.

Chain-of-thought (CoT) reasoning has been proposed as a transparency mechanism for large language models in safety-critical deployments, yet its effectiveness depends on faithfulness (whether models accurately verbalize the factors that actually influence their outputs), a property that prior evaluations have examined in only two proprietary models, finding acknowledgment rates as low as 25% for Claude 3.7 Sonnet and 39% for DeepSeek-R1. To extend this evaluation across the open-weight ecosystem, this study tests 12 open-weight reasoning models spanning 9 architectural families (7B-685B parameters) on 498 multiple-choice questions from MMLU and GPQA Diamond, injecting six categories of reasoning hints (sycophancy, consistency, visual pattern, metadata, grader hacking, and unethical information) and measuring the rate at which models acknowledge hint influence in their CoT when hints successfully alter answers. Across 41,832 inference runs, overall faithfulness rates range from 39.7% (Seed-1.6-Flash) to 89.9% (DeepSeek-V3.2-Speciale) across model families, with consistency hints (35.5%) and sycophancy hints (53.9%) exhibiting the lowest acknowledgment rates. Training methodology and model family predict faithfulness more strongly than parameter count, and keyword-based analysis reveals a striking gap between thinking-token acknowledgment (approximately 87.5%) and answer-text acknowledgment (approximately 28.6%), suggesting that models internally recognize hint influence but systematically suppress this acknowledgment in their outputs. These findings carry direct implications for the viability of CoT monitoring as a safety mechanism and suggest that faithfulness is not a fixed property of reasoning models but varies systematically with architecture, training method, and the nature of the influencing cue.

The inference overhead induced by redundant reasoning undermines the interactive experience and severely bottlenecks the deployment of Large Reasoning Models. Existing reinforcement learning (RL)-based solutions tackle this problem by coupling a length penalty with outcome-based rewards. This simplistic reward weighting struggles to reconcile brevity with accuracy, as enforcing brevity may compromise critical reasoning logic. In this work, we address this limitation by proposing a multi-agent RL framework that selectively penalizes redundant chunks, while preserving essential reasoning logic. Our framework, Self-Compression via MARL (SCMA), instantiates redundancy detection and evaluation through two specialized agents: a Segmentation Agent for decomposing the reasoning process into logical chunks, and a Scoring Agent for quantifying the significance of each chunk. The Segmentation and Scoring agents collaboratively define an importance-weighted length penalty during training, incentivizing a Reasoning Agent to prioritize essential logic without introducing inference overhead during deployment. Empirical evaluations across model scales demonstrate that SCMA reduces response length by 11.1\% to 39.0\% while boosting accuracy by 4.33\% to 10.02\%. Furthermore, ablation studies and qualitative analysis validate that the synergistic optimization within the MARL framework fosters emergent behaviors, yielding more powerful LRMs compared to vanilla RL paradigms.

Large language models (LLMs) are designed to perform a wide range of tasks. To improve their ability to solve complex problems requiring multi-step reasoning, recent research leverages process reward modeling to provide fine-grained feedback at each step of the reasoning process for reinforcement learning (RL), but it predominantly focuses on English. In this paper, we tackle the critical challenge of extending process reward models (PRMs) to multilingual settings. To achieve this, we train multilingual PRMs on a dataset spanning seven languages, which is translated from English. Through comprehensive evaluations on two widely used reasoning benchmarks across 11 languages, we demonstrate that multilingual PRMs not only improve average accuracy but also reduce early-stage reasoning errors. Furthermore, our results highlight the sensitivity of multilingual PRMs to both the number of training languages and the volume of English data, while also uncovering the benefits arising from more candidate responses and trainable parameters. This work opens promising avenues for robust multilingual applications in complex, multi-step reasoning tasks. In addition, we release the code to foster research along this line.

Conventional demographic inference methods have predominantly operated under the supervision of accurately labeled data, yet struggle to adapt to shifting social landscapes and diverse cultural contexts, leading to narrow specialization and limited accuracy in applications. Recently, the emergence of large multimodal models (LMMs) has shown transformative potential across various research tasks, such as visual comprehension and description. In this study, we explore the application of LMMs to demographic inference and introduce a benchmark for both quantitative and qualitative evaluation. Our findings indicate that LMMs possess advantages in zero-shot learning, interpretability, and handling uncurated 'in-the-wild' inputs, albeit with a propensity for off-target predictions. To enhance LMM performance and achieve comparability with supervised learning baselines, we propose a Chain-of-Thought augmented prompting approach, which effectively mitigates the off-target prediction issue.

Answering questions with Chain-of-Thought (CoT) has significantly enhanced the reasoning capabilities of Large Language Models (LLMs), yet its impact on Large Multimodal Models (LMMs) still lacks a systematic assessment and in-depth investigation. In this paper, we introduce MME-CoT, a specialized benchmark evaluating the CoT reasoning performance of LMMs, spanning six domains: math, science, OCR, logic, space-time, and general scenes. As the first comprehensive study in this area, we propose a thorough evaluation suite incorporating three novel metrics that assess the reasoning quality, robustness, and efficiency at a fine-grained level. Leveraging curated high-quality data and a unique evaluation strategy, we conduct an in-depth analysis of state-of-the-art LMMs, uncovering several key insights: 1) Models with reflection mechanism demonstrate a superior CoT quality, with Kimi k1.5 outperforming GPT-4o and demonstrating the highest quality results; 2) CoT prompting often degrades LMM performance on perception-heavy tasks, suggesting a potentially harmful overthinking behavior; and 3) Although the CoT quality is high, LMMs with reflection exhibit significant inefficiency in both normal response and self-correction phases. We hope MME-CoT serves as a foundation for advancing multimodal reasoning in LMMs. Project Page: https://mmecot.github.io/

Visual storytelling systems struggle to maintain character identity across frames and link actions to appropriate subjects, frequently leading to referential hallucinations. These issues can be addressed through grounding of characters, objects, and other entities on the visual elements. We propose StoryReasoning, a dataset containing 4,178 stories derived from 52,016 movie images, with both structured scene analyses and grounded stories. Each story maintains character and object consistency across frames while explicitly modeling multi-frame relationships through structured tabular representations. Our approach features cross-frame object re-identification using visual similarity and face recognition, chain-of-thought reasoning for explicit narrative modeling, and a grounding scheme that links textual elements to visual entities across multiple frames. We establish baseline performance by fine-tuning Qwen2.5-VL 7B, creating Qwen Storyteller, which performs end-to-end object detection, re-identification, and landmark detection while maintaining consistent object references throughout the story. Evaluation demonstrates a reduction from 4.06 to 3.56 (-12.3%) hallucinations on average per story when compared to a non-fine-tuned model.

Large language models (LLMs) have demonstrated strong reasoning abilities across specialized domains, motivating research into their application to legal reasoning. However, existing legal benchmarks often conflate factual recall with genuine inference, fragment the reasoning process, and overlook the quality of reasoning. To address these limitations, we introduce MSLR, the first Chinese multi-step legal reasoning dataset grounded in real-world judicial decision making. MSLR adopts the IRAC framework (Issue, Rule, Application, Conclusion) to model structured expert reasoning from official legal documents. In addition, we design a scalable Human-LLM collaborative annotation pipeline that efficiently produces fine-grained step-level reasoning annotations and provides a reusable methodological framework for multi-step reasoning datasets. Evaluation of multiple LLMs on MSLR shows only moderate performance, highlighting the challenges of adapting to complex legal reasoning. Further experiments demonstrate that Self-Initiated Chain-of-Thought prompts generated by models autonomously improve reasoning coherence and quality, outperforming human-designed prompts. MSLR contributes to advancing LLM reasoning and Chain-of-Thought strategies and offers open resources for future research. The dataset and code are available at https://github.com/yuwenhan07/MSLR-Bench and https://law.sjtu.edu.cn/flszyjzx/index.html.

Recent advances in large language models elicit reasoning in a chain of thought that allows models to decompose problems in a human-like fashion. Though this paradigm improves multi-step reasoning ability in language models, it is limited by being unimodal and applied mainly to question-answering tasks. We claim that incorporating visual augmentation into reasoning is essential, especially for complex, imaginative tasks. Consequently, we introduce VCoT, a novel method that leverages chain of thought prompting with vision-language grounding to recursively bridge the logical gaps within sequential data. Our method uses visual guidance to generate synthetic multimodal infillings that add consistent and novel information to reduce the logical gaps for downstream tasks that can benefit from temporal reasoning, as well as provide interpretability into models' multi-step reasoning. We apply VCoT to the Visual Storytelling and WikiHow summarization datasets and demonstrate through human evaluation that VCoT offers novel and consistent synthetic data augmentation beating chain of thought baselines, which can be used to enhance downstream performance.

Recently, Multimodal Large Language Models (MLLMs) have been widely integrated into diffusion frameworks primarily as text encoders to tackle complex tasks such as spatial reasoning. However, this paradigm suffers from two critical limitations: (i) MLLMs text encoder exhibits insufficient reasoning depth. Single-step encoding fails to activate the Chain-of-Thought process, which is essential for MLLMs to provide accurate guidance for complex tasks. (ii) The guidance remains invariant during the decoding process. Invariant guidance during decoding prevents DiT from progressively decomposing complex instructions into actionable denoising steps, even with correct MLLM encodings. To this end, we propose Endogenous Chain-of-Thought (EndoCoT), a novel framework that first activates MLLMs' reasoning potential by iteratively refining latent thought states through an iterative thought guidance module, and then bridges these states to the DiT's denoising process. Second, a terminal thought grounding module is applied to ensure the reasoning trajectory remains grounded in textual supervision by aligning the final state with ground-truth answers. With these two components, the MLLM text encoder delivers meticulously reasoned guidance, enabling the DiT to execute it progressively and ultimately solve complex tasks in a step-by-step manner. Extensive evaluations across diverse benchmarks (e.g., Maze, TSP, VSP, and Sudoku) achieve an average accuracy of 92.1%, outperforming the strongest baseline by 8.3 percentage points.

We present RALL-E, a robust language modeling method for text-to-speech (TTS) synthesis. While previous work based on large language models (LLMs) shows impressive performance on zero-shot TTS, such methods often suffer from poor robustness, such as unstable prosody (weird pitch and rhythm/duration) and a high word error rate (WER), due to the autoregressive prediction style of language models. The core idea behind RALL-E is chain-of-thought (CoT) prompting, which decomposes the task into simpler steps to enhance the robustness of LLM-based TTS. To accomplish this idea, RALL-E first predicts prosody features (pitch and duration) of the input text and uses them as intermediate conditions to predict speech tokens in a CoT style. Second, RALL-E utilizes the predicted duration prompt to guide the computing of self-attention weights in Transformer to enforce the model to focus on the corresponding phonemes and prosody features when predicting speech tokens. Results of comprehensive objective and subjective evaluations demonstrate that, compared to a powerful baseline method VALL-E, RALL-E significantly improves the WER of zero-shot TTS from 6.3% (without reranking) and 2.1% (with reranking) to 2.8% and 1.0%, respectively. Furthermore, we demonstrate that RALL-E correctly synthesizes sentences that are hard for VALL-E and reduces the error rate from 68% to 4%.

Recent advancements in large language models have showcased their remarkable generalizability across various domains. However, their reasoning abilities still have significant room for improvement, especially when confronted with scenarios requiring multi-step reasoning. Although large language models possess extensive knowledge, their behavior, particularly in terms of reasoning, often fails to effectively utilize this knowledge to establish a coherent thinking paradigm. Generative language models sometimes show hallucinations as their reasoning procedures are unconstrained by logical principles. Aiming to improve the zero-shot chain-of-thought reasoning ability of large language models, we propose Logical Chain-of-Thought (LogiCoT), a neurosymbolic framework that leverages principles from symbolic logic to verify and revise the reasoning processes accordingly. Experimental evaluations conducted on language tasks in diverse domains, including arithmetic, commonsense, symbolic, causal inference, and social problems, demonstrate the efficacy of the enhanced reasoning paradigm by logic.

Recent advancements in Vision-Language Models (VLMs) have significantly transformed various industries. In agriculture, these dual-modal capabilities offer promising applications such as precision farming, crop monitoring, pest detection, and environmental sustainability. While several Visual Question Answering (VQA) datasets and benchmarks have been developed to evaluate VLM performance, they often fail to adequately assess the critical reasoning and problem-solving skills required in complex agricultural contexts. To address this gap, we introduce AgriCoT, a VQA dataset that incorporates Chain-of-Thought (CoT) reasoning, specifically designed to evaluate the reasoning capabilities of VLMs. With 4,535 carefully curated samples, AgriCoT offers a comprehensive and robust evaluation of reasoning abilities for VLMs, particularly in zero-shot scenarios, by focusing on their capacity to engage in logical reasoning and effective problem-solving. Our evaluations, conducted with 26 representative VLMs, including both proprietary and open-source models, reveal that while some proprietary models excel at answering questions, there is a notable and significant gap in their reasoning capabilities. This underscores the importance of incorporating CoT for more precise and effective assessments. Our dataset are available at https://huggingface.co/datasets/wenyb/AgriCoT.

Large language models (LLMs) have shown significant promise in text reranking tasks by leveraging their advanced language understanding and reasoning capabilities. However, traditional supervised fine-tuning (SFT) approaches by ranking utilities can compromise LLMs' general-purpose abilities. To address this challenge, we propose a novel LLM-based reranking algorithm -- RaCT -- that implements SFT with Chain-of-Thought prompting, followed by a ranking preference optimization (RPO). The proposed RaCT aims to enhance ranking performance for LLMs while preserving their inherent language modeling abilities. Experimental evaluations on the three public ranking benchmarks (TREC DL, BEIR, and BRIGHT) and one LLM benchmark demonstrate the superior ranking performance of RaCT with a retained language understanding and reasoning capacity.

Artificial intelligence has advanced in Medical Visual Question Answering (Med-VQA), but prevalent research tends to focus on the accuracy of the answers, often overlooking the reasoning paths and interpretability, which are crucial in clinical settings. Besides, current Med-VQA algorithms, typically reliant on singular models, lack the robustness needed for real-world medical diagnostics which usually require collaborative expert evaluation. To address these shortcomings, this paper presents MedCoT, a novel hierarchical expert verification reasoning chain method designed to enhance interpretability and accuracy in biomedical imaging inquiries. MedCoT is predicated on two principles: The necessity for explicit reasoning paths in Med-VQA and the requirement for multi-expert review to formulate accurate conclusions. The methodology involves an Initial Specialist proposing diagnostic rationales, followed by a Follow-up Specialist who validates these rationales, and finally, a consensus is reached through a vote among a sparse Mixture of Experts within the locally deployed Diagnostic Specialist, which then provides the definitive diagnosis. Experimental evaluations on four standard Med-VQA datasets demonstrate that MedCoT surpasses existing state-of-the-art approaches, providing significant improvements in performance and interpretability.

Multi-step symbolic reasoning is critical for advancing downstream performance on financial tasks. Yet, benchmarks for systematically evaluating this capability are lacking. Existing datasets like FinQA and ConvFinQA supervise only final numerical answers, without assessing intermediate reasoning steps. To address this, we introduce FinChain, the first symbolic benchmark designed for verifiable Chain-of- Thought (CoT) financial reasoning. Spanning 54 topics across 12 financial domains, Fin- Chain offers five parameterized templates per topic, each varying in reasoning complexity and domain expertise required. Each dataset instance includes an executable Python trace, enabling automatic generation of extensive training data and easy adaptation to other domains. We also introduce ChainEval, a new metric for automatic evaluation of both final answers and intermediate reasoning. Benchmarking 30 LLMs on our dataset, we find that even state-of-the-art models have considerable room for improvement in multi-step financial reasoning. All templates and evaluation metrics for FinChain are available at https: //github.com/mbzuai-nlp/finchain.

Visual-Interleaved Chain-of-Thought (VI-CoT) enables MLLMs to continually update their understanding and decisions based on step-wise intermediate visual states (IVS), much like a human would, which demonstrates impressive success in various tasks, thereby leading to emerged advancements in related benchmarks. Despite promising progress, current benchmarks provide models with relatively fixed IVS, rather than free-style IVS, whch might forcibly distort the original thinking trajectories, failing to evaluate their intrinsic reasoning capabilities. More importantly, existing benchmarks neglect to systematically explore the impact factors that IVS would impart to untamed reasoning performance. To tackle above gaps, we introduce a specialized benchmark termed ViC-Bench, consisting of four representive tasks: maze navigation, jigsaw puzzle, embodied long-horizon planning, and complex counting, where each task has dedicated free-style IVS generation pipeline supporting function calls. To systematically examine VI-CoT capability, we propose a thorough evaluation suite incorporating a progressive three-stage strategy with targeted new metrics. Besides, we establish Incremental Prompting Information Injection (IPII) strategy to ablatively explore the prompting factors for VI-CoT. We extensively conduct evaluations for 18 advanced MLLMs, revealing key insights into their VI-CoT capability. Our proposed benchmark is publicly open at Huggingface.

Although large language models (LLMs) have achieved excellent performance in a variety of evaluation benchmarks, they still struggle in complex reasoning tasks which require specific knowledge and multi-hop reasoning. To improve the reasoning abilities, we propose ChatCoT, a tool-augmented chain-of-thought reasoning framework for chat-based LLMs. In ChatCoT, we model the chain-of-thought~(CoT) reasoning as multi-turn conversations, to utilize tools in a more natural way through chatting. At each turn, LLMs can either interact with tools or perform the reasoning. Our approach can effectively leverage the multi-turn conversation ability of chat-based LLMs, and integrate the thought chain following and tools manipulation in a unified way. Specially, we initialize the early turns of the conversation by the tools, tasks and reasoning format, and propose an iterative tool-augmented reasoning step to perform step-by-step tool-augmented reasoning. The experiment results on two complex reasoning datasets (MATH and HotpotQA) have shown the effectiveness of ChatCoT on complex reasoning tasks, achieving a 6.8\% relative improvement over the state-of-the-art baseline. Our code and data are available at: https://github.com/RUCAIBOX/ChatCoT.

While the recent Chain-of-Thought (CoT) technique enhances the reasoning ability of large language models (LLMs) with the theory of mind, it might still struggle in handling logical reasoning that relies much on symbolic expressions and rigid deducing rules. To strengthen the logical reasoning capability of LLMs, we propose a novel Symbolic Chain-of-Thought, namely SymbCoT, a fully LLM-based framework that integrates symbolic expressions and logic rules with CoT prompting. Technically, building upon an LLM, SymbCoT 1) first translates the natural language context into the symbolic format, and then 2) derives a step-by-step plan to solve the problem with symbolic logical rules, 3) followed by a verifier to check the translation and reasoning chain. Via thorough evaluations on 5 standard datasets with both First-Order Logic and Constraint Optimization symbolic expressions, SymbCoT shows striking improvements over the CoT method consistently, meanwhile refreshing the current state-of-the-art performances. We further demonstrate that our system advances in more faithful, flexible, and explainable logical reasoning. To our knowledge, this is the first to combine symbolic expressions and rules into CoT for logical reasoning with LLMs. Code is open at https://github.com/Aiden0526/SymbCoT.

Chain-of-thought (CoT) is capable of eliciting models to explicitly generate reasoning paths, thus promoting reasoning accuracy and attracting increasing attention. Specifically, zero-shot CoT achieves remarkable improvements in a wide range of reasoning tasks by simply instructing the LLM with the prompt "Let's think step by step!". Despite the success of zero-shot CoT, the existing zero-shot prompting techniques remain limited to a single language, making it challenging to generalize to other languages and hindering global development. In this work, we introduce cross-lingual prompting (CLP), aiming to improve zero-shot CoT reasoning across languages. Specifically, CLP consists of two main components: (1) cross-lingual alignment prompting and (2) task-specific solver prompting. The cross-lingual alignment prompting is responsible for aligning representations across different languages, whereas the task-specific solver prompting is used to generate the final chain of thoughts and results for the reasoning task. In addition, we further introduce cross-lingual self-consistent prompting (CLSP) to ensemble different reasoning paths across languages. Our experimental evaluations on several benchmarks demonstrate that CLP and CLSP significantly outperform the existing prompting methods and achieve state-of-the-art performance. We hope this work will inspire further breakthroughs in cross-lingual CoT.

We identify two crucial limitations in the evaluation of recent parallel-integrated method Parallel Context Windows (PCW), which extends the maximum context lengths of language models, e.g., 2048 for LLaMA, by harnessing window-wise attention and positional embedding techniques. We first show that a simple yet strong baseline, weighted sum ensemble, is missing for the in-context few-shot classification. Moreover, on more challenging Chain-of-Thought (CoT) reasoning (e.g., HotpotQA), PCW would present unexpected deterioration regarding question miscomprehension and false inference. Based on our findings, we suggest that the existing PCW design may not guarantee sufficient improvement and practicality in handling lengthy documents in real-world applications. More community efforts on enabling language models' long context understanding ability should be paid.

We introduce Skywork R1V, a multimodal reasoning model extending the an R1-series Large language models (LLM) to visual modalities via an efficient multimodal transfer method. Leveraging a lightweight visual projector, Skywork R1V facilitates seamless multimodal adaptation without necessitating retraining of either the foundational language model or the vision encoder. To strengthen visual-text alignment, we propose a hybrid optimization strategy that combines Iterative Supervised Fine-Tuning (SFT) with Group Relative Policy Optimization (GRPO), significantly enhancing cross-modal integration efficiency. Additionally, we introduce an adaptive-length Chain-of-Thought distillation approach for reasoning data generation. This approach dynamically optimizes reasoning chain lengths, thereby enhancing inference efficiency and preventing excessive reasoning overthinking. Empirical evaluations demonstrate that Skywork R1V, with only 38B parameters, delivers competitive performance, achieving a score of 69.0 on the MMMU benchmark and 67.5 on MathVista. Meanwhile, it maintains robust textual reasoning performance, evidenced by impressive scores of 72.0 on AIME and 94.0 on MATH500. The Skywork R1V model weights have been publicly released to promote openness and reproducibility.

Recent advances in multimodal large language models (MLLMs) have demonstrated remarkable capabilities in vision-language tasks, yet they often struggle with vision-centric scenarios where precise visual focus is needed for accurate reasoning. In this paper, we introduce Argus to address these limitations with a new visual attention grounding mechanism. Our approach employs object-centric grounding as visual chain-of-thought signals, enabling more effective goal-conditioned visual attention during multimodal reasoning tasks. Evaluations on diverse benchmarks demonstrate that Argus excels in both multimodal reasoning tasks and referring object grounding tasks. Extensive analysis further validates various design choices of Argus, and reveals the effectiveness of explicit language-guided visual region-of-interest engagement in MLLMs, highlighting the importance of advancing multimodal intelligence from a visual-centric perspective. Project page: https://yunzeman.github.io/argus/

In the domain of text-to-image generative models, biases inherent in training datasets often propagate into generated content, posing significant ethical challenges, particularly in socially sensitive contexts. We introduce FairCoT, a novel framework that enhances fairness in diffusion models through Chain-of-Thought (CoT) reasoning within multimodal generative large language models (LLMs). FairCoT employs iterative CoT refinement and attire-based attribute prediction to systematically mitigate biases, ensuring diverse and equitable representation in generated images. By integrating iterative reasoning processes, FairCoT addresses the limitations of zero-shot CoT in sensitive scenarios, balancing creativity with ethical responsibility. Experimental evaluations across multiple models, including DALL-E and various Stable Diffusion variants, demonstrate that FairCoT significantly improves fairness and diversity metrics without compromising image quality or relevance. Our approach advances ethical AI practices in generative modeling, promoting socially responsible content generation and setting new standards for fairness in AI-generated imagery.

Large reasoning models (LRMs) have achieved impressive performance in complex tasks, often outperforming conventional large language models (LLMs). However, the prevalent issue of overthinking severely limits their computational efficiency. Overthinking occurs when models generate excessive and redundant tokens that contribute little to accurate outcomes, especially in simple tasks, resulting in a significant waste of computational resources. To systematically investigate this issue, we introduce Think-Bench, a benchmark designed to evaluate the reasoning efficiency of LRMs. We also propose novel efficiency metrics and conduct a comprehensive evaluation of various LRMs across multiple dimensions, including the reasoning process, outcome quality, and chain-of-thought (CoT) characteristics. Our analysis reveals that most LRMs exhibit overthinking in handling easy questions, generating unnecessarily lengthy reasoning chains. While many LRMs demonstrate high CoT quality, several suffer from low efficiency. We hope that Think-Bench can serve as a robust foundation for advancing research into LRMs.

Language instructions and demonstrations are two natural ways for users to teach robots personalized tasks. Recent progress in Large Language Models (LLMs) has shown impressive performance in translating language instructions into code for robotic tasks. However, translating demonstrations into task code continues to be a challenge due to the length and complexity of both demonstrations and code, making learning a direct mapping intractable. This paper presents Demo2Code, a novel framework that generates robot task code from demonstrations via an extended chain-of-thought and defines a common latent specification to connect the two. Our framework employs a robust two-stage process: (1) a recursive summarization technique that condenses demonstrations into concise specifications, and (2) a code synthesis approach that expands each function recursively from the generated specifications. We conduct extensive evaluation on various robot task benchmarks, including a novel game benchmark Robotouille, designed to simulate diverse cooking tasks in a kitchen environment. The project's website is available at https://portal-cornell.github.io/demo2code/

Chain-of-thought (CoT) prompting has become a widely used strategy for working with large language and multimodal models. While CoT has been shown to improve performance across many tasks, determining the settings in which it is effective remains an ongoing effort. In particular, it is still an open question in what settings CoT systematically reduces model performance. In this paper, we seek to identify the characteristics of tasks where CoT reduces performance by drawing inspiration from cognitive psychology, looking at cases where (i) verbal thinking or deliberation hurts performance in humans, and (ii) the constraints governing human performance generalize to language models. Three such cases are implicit statistical learning, visual recognition, and classifying with patterns containing exceptions. In extensive experiments across all three settings, we find that a diverse collection of state-of-the-art models exhibit significant drop-offs in performance (e.g., up to 36.3% absolute accuracy for OpenAI o1-preview compared to GPT-4o) when using inference-time reasoning compared to zero-shot counterparts. We also identify three tasks that satisfy condition (i) but not (ii), and find that while verbal thinking reduces human performance in these tasks, CoT retains or increases model performance. Overall, our results show that while there is not an exact parallel between the cognitive processes of models and those of humans, considering cases where thinking has negative consequences for human performance can help us identify settings where it negatively impacts models. By connecting the literature on human deliberation with evaluations of CoT, we offer a new tool that can be used in understanding the impact of prompt choices and inference-time reasoning.

Reinforcement learning (RL) fine-tuning has become a key technique for enhancing large language models (LLMs) on reasoning-intensive tasks, motivating its extension to vision language models (VLMs). While RL-tuned VLMs improve on visual reasoning benchmarks, they remain vulnerable to weak visual grounding, hallucinations, and over-reliance on textual cues. We show that simple, controlled textual perturbations--misleading captions or incorrect chain-of-thought (CoT) traces--cause substantial drops in robustness and confidence, and that these effects are more pronounced when CoT consistency is taken into account across open-source multimodal reasoning models. Entropy-based metrics further show that these perturbations reshape model uncertainty and probability mass on the correct option, exposing model-specific trends in miscalibration. To better understand these vulnerabilities, we further analyze RL fine-tuning dynamics and uncover an accuracy-faithfulness trade-off: fine-tuning raises benchmark accuracy, but can simultaneously erode the reliability of the accompanying CoT and its robustness to contextual shifts. Although adversarial augmentation improves robustness, it does not by itself prevent faithfulness drift. Incorporating a faithfulness-aware reward can restore alignment between answers and reasoning, but when paired with augmentation, training risks collapsing onto shortcut strategies and robustness remains elusive. Together, these findings highlight the limitations of accuracy-only evaluations and motivate training and assessment protocols that jointly emphasize correctness, robustness, and the faithfulness of visually grounded reasoning.

Trajectory-based motion control has emerged as an intuitive and efficient approach for controllable video generation. However, the existing trajectory-based approaches are usually limited to only generating the motion trajectory of the controlled object and ignoring the dynamic interactions between the controlled object and its surroundings. To address this limitation, we propose a Chain-of-Thought-based motion controller for controllable video generation, named C-Drag. Instead of directly generating the motion of some objects, our C-Drag first performs object perception and then reasons the dynamic interactions between different objects according to the given motion control of the objects. Specifically, our method includes an object perception module and a Chain-of-Thought-based motion reasoning module. The object perception module employs visual language models to capture the position and category information of various objects within the image. The Chain-of-Thought-based motion reasoning module takes this information as input and conducts a stage-wise reasoning process to generate motion trajectories for each of the affected objects, which are subsequently fed to the diffusion model for video synthesis. Furthermore, we introduce a new video object interaction (VOI) dataset to evaluate the generation quality of motion controlled video generation methods. Our VOI dataset contains three typical types of interactions and provides the motion trajectories of objects that can be used for accurate performance evaluation. Experimental results show that C-Drag achieves promising performance across multiple metrics, excelling in object motion control. Our benchmark, codes, and models will be available at https://github.com/WesLee88524/C-Drag-Official-Repo.

BIG-Bench (Srivastava et al., 2022) is a diverse evaluation suite that focuses on tasks believed to be beyond the capabilities of current language models. Language models have already made good progress on this benchmark, with the best model in the BIG-Bench paper outperforming average reported human-rater results on 65% of the BIG-Bench tasks via few-shot prompting. But on what tasks do language models fall short of average human-rater performance, and are those tasks actually unsolvable by current language models? In this work, we focus on a suite of 23 challenging BIG-Bench tasks which we call BIG-Bench Hard (BBH). These are the task for which prior language model evaluations did not outperform the average human-rater. We find that applying chain-of-thought (CoT) prompting to BBH tasks enables PaLM to surpass the average human-rater performance on 10 of the 23 tasks, and Codex (code-davinci-002) to surpass the average human-rater performance on 17 of the 23 tasks. Since many tasks in BBH require multi-step reasoning, few-shot prompting without CoT, as done in the BIG-Bench evaluations (Srivastava et al., 2022), substantially underestimates the best performance and capabilities of language models, which is better captured via CoT prompting. As further analysis, we explore the interaction between CoT and model scale on BBH, finding that CoT enables emergent task performance on several BBH tasks with otherwise flat scaling curves.

Large language models (LLMs) have shown impressive multilingual capabilities through pretraining on diverse corpora. While these models show strong reasoning abilities, their performance varies significantly across languages due to uneven training data distribution. Existing approaches using machine translation, and extensive multilingual pretraining and cross-lingual tuning face scalability challenges and often fail to capture nuanced reasoning processes across languages. In this paper, we introduce AdaCoT (Adaptive Chain-of-Thought), a framework that enhances multilingual reasoning by dynamically routing thought processes through intermediary "thinking languages" before generating target-language responses. AdaCoT leverages a language-agnostic core and incorporates an adaptive, reward-based mechanism for selecting optimal reasoning pathways without requiring additional pretraining. Our comprehensive evaluation across multiple benchmarks demonstrates substantial improvements in both factual reasoning quality and cross-lingual consistency, with particularly strong performance gains in low-resource language settings. The results suggest that adaptive reasoning paths can effectively bridge the performance gap between high and low-resource languages while maintaining cultural and linguistic nuances.

Data visualization serves as a critical means for presenting data and mining its valuable insights. The task of chart summarization, through natural language processing techniques, facilitates in-depth data analysis of charts. However, there still are notable deficiencies in terms of visual-language matching and reasoning ability for existing approaches. To address these limitations, this study constructs a large-scale dataset of comprehensive chart-caption pairs and fine-tuning instructions on each chart. Thanks to the broad coverage of various topics and visual styles within this dataset, better matching degree can be achieved from the view of training data. Moreover, we propose an innovative chart summarization method, ChartThinker, which synthesizes deep analysis based on chains of thought and strategies of context retrieval, aiming to improve the logical coherence and accuracy of the generated summaries. Built upon the curated datasets, our trained model consistently exhibits superior performance in chart summarization tasks, surpassing 8 state-of-the-art models over 7 evaluation metrics. Our dataset and codes are publicly accessible.

Chain-of-Thought (CoT) reasoning has proven effective in enhancing large language models by encouraging step-by-step intermediate reasoning, and recent advances have extended this paradigm to Multimodal Large Language Models (MLLMs). In the medical domain, where diagnostic decisions depend on nuanced visual cues and sequential reasoning, CoT aligns naturally with clinical thinking processes. However, Current benchmarks for medical image understanding generally focus on the final answer while ignoring the reasoning path. An opaque process lacks reliable bases for judgment, making it difficult to assist doctors in diagnosis. To address this gap, we introduce a new M3CoTBench benchmark specifically designed to evaluate the correctness, efficiency, impact, and consistency of CoT reasoning in medical image understanding. M3CoTBench features 1) a diverse, multi-level difficulty dataset covering 24 examination types, 2) 13 varying-difficulty tasks, 3) a suite of CoT-specific evaluation metrics (correctness, efficiency, impact, and consistency) tailored to clinical reasoning, and 4) a performance analysis of multiple MLLMs. M3CoTBench systematically evaluates CoT reasoning across diverse medical imaging tasks, revealing current limitations of MLLMs in generating reliable and clinically interpretable reasoning, and aims to foster the development of transparent, trustworthy, and diagnostically accurate AI systems for healthcare. Project page at https://juntaojianggavin.github.io/projects/M3CoTBench/.

In this paper, we propose a method for generating layouts for image-based advertisements by leveraging a Vision-Language Model (VLM). Conventional advertisement layout techniques have predominantly relied on saliency mapping to detect salient regions within a background image, but such approaches often fail to fully account for the image's detailed composition and semantic content. To overcome this limitation, our method harnesses a VLM to recognize the products and other elements depicted in the background and to inform the placement of text and logos. The proposed layout-generation pipeline consists of two steps. In the first step, the VLM analyzes the image to identify object types and their spatial relationships, then produces a text-based "placement plan" based on this analysis. In the second step, that plan is rendered into the final layout by generating HTML-format code. We validated the effectiveness of our approach through evaluation experiments, conducting both quantitative and qualitative comparisons against existing methods. The results demonstrate that by explicitly considering the background image's content, our method produces noticeably higher-quality advertisement layouts.

Recently, there have been notable advancements in large language models (LLMs), demonstrating their growing abilities in complex reasoning. However, existing research largely overlooks a thorough and systematic comparison of these models' reasoning processes and outputs, particularly regarding their self-reflection pattern (also termed "Aha moment") and the interconnections across diverse domains. This paper proposes a novel framework for analyzing the reasoning characteristics of four cutting-edge large reasoning models (GPT-o1, DeepSeek-R1, Kimi-k1.5, and Grok-3) using keywords statistic and LLM-as-a-judge paradigm. Our approach connects their internal thinking processes with their final outputs. A diverse dataset consists of real-world scenario-based questions covering logical deduction, causal inference, and multi-step problem-solving. Additionally, a set of metrics is put forward to assess both the coherence of reasoning and the accuracy of the outputs. The research results uncover various patterns of how these models balance exploration and exploitation, deal with problems, and reach conclusions during the reasoning process. Through quantitative and qualitative comparisons, disparities among these models are identified in aspects such as the depth of reasoning, the reliance on intermediate steps, and the degree of similarity between their thinking processes and output patterns and those of GPT-o1. This work offers valuable insights into the trade-off between computational efficiency and reasoning robustness and provides practical recommendations for enhancing model design and evaluation in practical applications. We publicly release our project at: https://github.com/ChangWenhan/FromThinking2Output

In the rapidly evolving field of Natural Language Processing, Large Language Models (LLMs) are tasked with increasingly complex reasoning challenges. Traditional methods like chain-of-thought prompting have shown promise but often fall short in fully leveraging a model's reasoning capabilities. This paper introduces SQuARE (Sequential Question Answering Reasoning Engine), a novel prompting technique designed to improve reasoning through a self-interrogation paradigm. Building upon CoT frameworks, SQuARE prompts models to generate and resolve multiple auxiliary questions before tackling the main query, promoting a more thorough exploration of various aspects of a topic. Our expansive evaluations, conducted with Llama 3 and GPT-4o models across multiple question-answering datasets, demonstrate that SQuARE significantly surpasses traditional CoT prompts and existing rephrase-and-respond methods. By systematically decomposing queries, SQuARE advances LLM capabilities in reasoning tasks. The code is publicly available at https://github.com/IntelLabs/RAG-FiT/tree/square.

Leading large language models have demonstrated impressive capabilities in reasoning-intensive tasks, such as standardized educational testing. However, they often require extensive training in low-resource settings with inaccessible infrastructure. Small or compact models, though more efficient, frequently lack sufficient support for underrepresented languages, leaving a performance gap in critical domains. This work explores the potential of parameter-efficient fine-tuning of compact open-weight language models to handle reasoning-intensive tasks in the underrepresented Ukrainian language, building on the findings of the ZNO-Eval benchmark. Parameter-efficient fine-tuning of LLaMA 3.1 (8 billion parameters), LLaMA 3.2 (3 billion parameters), and Gemma 2 (9 billion parameters) models on chain-of-thought solutions resulted in a modest test score improvement of up to 17.4% on complex matching tasks and 1.6% overall compared to tuning on answer letters alone, offering enhanced interpretability and robustness. In addition, the proposed tuning method with joint task topic and step-by-step solution generation outperforms standard chain-of-thought tuning in matching tasks and provides a 5.4% gain over the best LLaMA 3.2 model due to guiding the model to recall and apply domain-relevant information. Contrasting obtained results with zero-shot evaluations of leading open-weight and proprietary models such as Qwen, DeepSeek R1, OpenAI o1 and o3, Gemini, and Claude, highlight that fine-tuning LLaMA and Gemma models with 2,032 step-by-step solutions and 20 to 50 million trainable parameters on a single A100 GPU lets them outperform GPT-4o mini, Mistral Large, and larger open-weight models. This research also evaluates how merging the quantized adapter with the base model influences the generation quality. Source code and tuned models are available at https://github.com/NLPForUA/ZNO.

WebAssembly enables near-native execution in web applications and is increasingly adopted for tasks that demand high performance and robust security. However, its assembly-like syntax, implicit stack machine, and low-level data types make it extremely difficult for human developers to understand, spurring the need for effective WebAssembly reverse engineering techniques. In this paper, we propose StackSight, a novel neurosymbolic approach that combines Large Language Models (LLMs) with advanced program analysis to decompile complex WebAssembly code into readable C++ snippets. StackSight visualizes and tracks virtual stack alterations via a static analysis algorithm and then applies chain-of-thought prompting to harness LLM's complex reasoning capabilities. Evaluation results show that StackSight significantly improves WebAssembly decompilation. Our user study also demonstrates that code snippets generated by StackSight have significantly higher win rates and enable a better grasp of code semantics.

With the rapid advancement of Large Language Models (LLMs), developing effective critic modules for precise guidance has become crucial yet challenging. In this paper, we initially demonstrate that supervised fine-tuning for building critic modules (which is widely adopted in current solutions) fails to genuinely enhance models' critique abilities, producing superficial critiques with insufficient reflections and verifications. To unlock the unprecedented critique capabilities, we propose RefCritic, a long-chain-of-thought critic module based on reinforcement learning with dual rule-based rewards: (1) instance-level correctness of solution judgments and (2) refinement accuracies of the policy model based on critiques, aiming to generate high-quality evaluations with actionable feedback that effectively guides model refinement. We evaluate RefCritic on Qwen2.5-14B-Instruct and DeepSeek-R1-Distill-Qwen-14B across five benchmarks. On critique and refinement settings, RefCritic demonstrates consistent advantages across all benchmarks, e.g., 6.8\% and 7.2\% gains on AIME25 for the respective base models. Notably, under majority voting, policy models filtered by RefCritic show superior scaling with increased voting numbers. Moreover, despite training on solution-level supervision, RefCritic outperforms step-level supervised approaches on ProcessBench, a benchmark to identify erroneous steps in mathematical reasoning.

Object referring aims to detect all objects in an image that match a given natural language description. We argue that a robust object referring model should be grounded, meaning its predictions should be both explainable and faithful to the visual content. Specifically, it should satisfy two key properties: 1) Verifiable, by producing interpretable reasoning that justifies its predictions and clearly links them to visual evidence; and 2) Trustworthy, by learning to abstain when no object in the image satisfies the given expression. However, most methods treat referring as a direct bounding box prediction task, offering limited interpretability and struggling to reject expressions with no matching object. In this work, we propose Rex-Thinker, a model that formulates object referring as an explicit CoT reasoning task. Given a referring expression, we first identify all candidate object instances corresponding to the referred object category. Rex-Thinker then performs step-by-step reasoning over each candidate to assess whether it matches the given expression, before making a final prediction. To support this paradigm, we construct a large-scale CoT-style referring dataset named HumanRef-CoT by prompting GPT-4o on the HumanRef dataset. Each reasoning trace follows a structured planning, action, and summarization format, enabling the model to learn decomposed, interpretable reasoning over object candidates. We then train Rex-Thinker in two stages: a cold-start supervised fine-tuning phase to teach the model how to perform structured reasoning, followed by GRPO-based RL learning to improve accuracy and generalization. Experiments show that our approach outperforms standard baselines in both precision and interpretability on in-domain evaluation, while also demonstrating improved ability to reject hallucinated outputs and strong generalization in out-of-domain settings.

Solving Bengali Math Word Problems (MWPs) remains a major challenge in natural language processing (NLP) due to the language's low-resource status and the multi-step reasoning required. Existing models struggle with complex Bengali MWPs, largely because no human-annotated Bengali dataset has previously addressed this task. This gap has limited progress in Bengali mathematical reasoning. To address this, we created SOMADHAN, a dataset of 8792 complex Bengali MWPs with manually written, step-by-step solutions. We designed this dataset to support reasoning-focused evaluation and model development in a linguistically underrepresented context. Using SOMADHAN, we evaluated a range of large language models (LLMs) - including GPT-4o, GPT-3.5 Turbo, LLaMA series models, Deepseek, and Qwen - through both zero-shot and few-shot prompting with and without Chain of Thought (CoT) reasoning. CoT prompting consistently improved performance over standard prompting, especially in tasks requiring multi-step logic. LLaMA-3.3 70B achieved the highest accuracy of 88% with few-shot CoT prompting. We also applied Low-Rank Adaptation (LoRA) to fine-tune models efficiently, enabling them to adapt to Bengali MWPs with minimal computational cost. Our work fills a critical gap in Bengali NLP by providing a high-quality reasoning dataset and a scalable framework for solving complex MWPs. We aim to advance equitable research in low-resource languages and enhance reasoning capabilities in educational and language technologies.

We propose MIRA, a new benchmark designed to evaluate models in scenarios where generating intermediate visual images is essential for successful reasoning. Unlike traditional CoT methods that rely solely on text, tasks in MIRA require models to generate and utilize intermediate images - such as sketches, structural diagrams, or path drawings - to guide their reasoning process. This setup closely mirrors how humans solve complex problems through "drawing to think". To solve this, MIRA focuses on tasks that are intrinsically challenging and involve complex structures, spatial relationships, or reasoning steps that are difficult to express through language alone. To ensure that our evaluation data is of high-quality, we include 546 multimodal problems, annotated with intermediate visual images and final answers. We also propose a unified evaluation protocol for MIRA that spans three levels of evaluation input: direct input with image and question only, text-only CoT input with image and thinking prompts, and Visual-CoT input with both annotated image clues and textual thinking prompts. To probe the upper bound of model capacity on our benchmark, we also report pass@k and majority voting accuracies under different k settings. Experimental results show that existing multimodal large language models, including strongest private models as well as strong open-weight models, perform poorly when relying solely on textual prompts. However, when intermediate visual cues are provided, model performance improves consistently, yielding an average relative gain of 33.7% across all models and tasks. We also probe the upper bound by expanding the search space and designing textual prompts aligned with Visual-CoT, but both yield only limited improvements compared to our Visual-CoT setting. These results underscore the critical role of imagined visual information in enabling successful reasoning on MIRA.

Recent frontier models employ long chain-of-thought reasoning to explore solution spaces in context and achieve stonger performance. While many works study distillation to build smaller yet capable models, most focus on English and little is known about language-specific reasoning. To bridge this gap, we first introduct **Language-Mixed CoT**, a reasoning schema that switches between English and a target language, using English as an anchor to excel in reasoning while minimizing translation artificats. As a Korean case study, we curate **Yi-Sang**: 5.79M native-Korean prompts from web Q&A, exams, STEM, and code; 3.7M long reasoning traces generated from Qwen3-32B; and a targeted 260k high-yield subset. We train ninve models (4B-35B) across six families (Qwen2.5, Llama-3.1, Gemma-3, etc). Our best model, **KO-REAson-35B**, achieves state-of-the-art performance, with the highest overall average score (64.0 \pm 25), ranking first on 5/9 benchmarks and second on the remainder. Samller and mid-sized models also benefit substantially, with an average improvement of +18.6 points across teh evaluated nine benchmarks. Ablations show **Language-Mixed CoT** is more effective than monolingual CoT, also resulting in cross-lingual and mult-modal performance gains. We release our data-curation pipeline, evaluation system, datasets, and models to advance research on language-specific reasoning. Data and model collection: https://huggingface.co/KOREAson.

Optical Chemical Structure Recognition (OCSR) is crucial for digitizing chemical knowledge by converting molecular images into machine-readable formats. While recent vision-language models (VLMs) have shown potential in this task, their image-captioning approach often struggles with complex molecular structures and inconsistent annotations. To overcome these challenges, we introduce GTR-Mol-VLM, a novel framework featuring two key innovations: (1) the Graph Traversal as Visual Chain of Thought mechanism that emulates human reasoning by incrementally parsing molecular graphs through sequential atom-bond predictions, and (2) the data-centric principle of Faithfully Recognize What You've Seen, which addresses the mismatch between abbreviated structures in images and their expanded annotations. To support model development, we constructed GTR-CoT-1.3M, a large-scale instruction-tuning dataset with meticulously corrected annotations, and introduced MolRec-Bench, the first benchmark designed for a fine-grained evaluation of graph-parsing accuracy in OCSR. Comprehensive experiments demonstrate that GTR-Mol-VLM achieves superior results compared to specialist models, chemistry-domain VLMs, and commercial general-purpose VLMs. Notably, in scenarios involving molecular images with functional group abbreviations, GTR-Mol-VLM outperforms the second-best baseline by approximately 14 percentage points, both in SMILES-based and graph-based metrics. We hope that this work will drive OCSR technology to more effectively meet real-world needs, thereby advancing the fields of cheminformatics and AI for Science. We will release GTR-CoT at https://github.com/opendatalab/GTR-CoT.

Computer-Aided Design (CAD) plays a vital role in engineering and manufacturing, yet current CAD workflows require extensive domain expertise and manual modeling effort. Recent advances in large language models (LLMs) have made it possible to generate code from natural language, opening new opportunities for automating parametric 3D modeling. However, directly translating human design intent into executable CAD code remains highly challenging, due to the need for logical reasoning, syntactic correctness, and numerical precision. In this work, we propose CAD-RL, a multimodal Chain-of-Thought (CoT) guided reinforcement learning post training framework for CAD modeling code generation. Our method combines CoT-based Cold Start with goal-driven reinforcement learning post training using three task-specific rewards: executability reward, geometric accuracy reward, and external evaluation reward. To ensure stable policy learning under sparse and high-variance reward conditions, we introduce three targeted optimization strategies: Trust Region Stretch for improved exploration, Precision Token Loss for enhanced dimensions parameter accuracy, and Overlong Filtering to reduce noisy supervision. To support training and benchmarking, we release ExeCAD, a noval dataset comprising 16,540 real-world CAD examples with paired natural language and structured design language descriptions, executable CADQuery scripts, and rendered 3D models. Experiments demonstrate that CAD-RL achieves significant improvements in reasoning quality, output precision, and code executability over existing VLMs.

Instruction generation is a vital and multidisciplinary research area with broad applications. Existing instruction generation models are limited to generating instructions in a single style from a particular dataset, and the style and content of generated instructions cannot be controlled. Moreover, most existing instruction generation methods also disregard the spatial modeling of the navigation environment. Leveraging the capabilities of Large Language Models (LLMs), we propose C-Instructor, which utilizes the chain-of-thought-style prompt for style-controllable and content-controllable instruction generation. Firstly, we propose a Chain of Thought with Landmarks (CoTL) mechanism, which guides the LLM to identify key landmarks and then generate complete instructions. CoTL renders generated instructions more accessible to follow and offers greater controllability over the manipulation of landmark objects. Furthermore, we present a Spatial Topology Modeling Task to facilitate the understanding of the spatial structure of the environment. Finally, we introduce a Style-Mixed Training policy, harnessing the prior knowledge of LLMs to enable style control for instruction generation based on different prompts within a single model instance. Extensive experiments demonstrate that instructions generated by C-Instructor outperform those generated by previous methods in text metrics, navigation guidance evaluation, and user studies.

Multi-modal Chain-of-Thought (MCoT) requires models to leverage knowledge from both textual and visual modalities for step-by-step reasoning, which gains increasing attention. Nevertheless, the current MCoT benchmark still faces some challenges: (1) absence of visual modal reasoning, (2) single-step visual modal reasoning, and (3) Domain missing, thereby hindering the development of MCoT. Motivated by this, we introduce a novel benchmark (M^3CoT) to address the above challenges, advancing the multi-domain, multi-step, and multi-modal CoT. Additionally, we conduct a thorough evaluation involving abundant MCoT approaches on Vision Large Language Models (VLLMs). In addition, we highlight that the current VLLMs still struggle to correctly reason in M^3CoT and there remains a large gap between existing VLLMs and human performance in M^3CoT, despite their superior results on previous MCoT benchmarks. To our knowledge, we take the first meaningful step toward the multi-domain, multi-step, and multi-modal scenario in MCoT. We hope that M^3CoT can serve as a valuable resource, providing a pioneering foundation in multi-domain, multi-step, multi-modal chain-of-thought research.

Recently, o1-like models have drawn significant attention, where these models produce the long Chain-of-Thought (CoT) reasoning steps to improve the reasoning abilities of existing Large Language Models (LLMs). In this paper, to understand the qualities of these long CoTs and measure the critique abilities of existing LLMs on these long CoTs, we introduce the DeltaBench, including the generated long CoTs from different o1-like models (e.g., QwQ, DeepSeek-R1) for different reasoning tasks (e.g., Math, Code, General Reasoning), to measure the ability to detect errors in long CoT reasoning. Based on DeltaBench, we first perform fine-grained analysis of the generated long CoTs to discover the effectiveness and efficiency of different o1-like models. Then, we conduct extensive evaluations of existing process reward models (PRMs) and critic models to detect the errors of each annotated process, which aims to investigate the boundaries and limitations of existing PRMs and critic models. Finally, we hope that DeltaBench could guide developers to better understand the long CoT reasoning abilities of their models.

Predicting wildfire risk is a reasoning-intensive spatial problem that requires the integration of visual, climatic, and geographic factors to infer continuous risk maps. Existing methods lack the causal reasoning and multimodal understanding required for reliable generalization. We introduce FireScope-Bench, a large-scale dataset and benchmark that couples Sentinel-2 imagery and climate data with expert-defined risk rasters across the USA, and real wildfire events in Europe for cross-continental evaluation. Building on this dataset, we propose FireScope, a VLM-based reasoning-to-generation framework that learns from both reinforcement learning and visual supervision to predict risk rasters with complementary reasoning traces. When trained in the USA and tested in Europe, FireScope achieves substantial performance gains, while expert feedback and automated analysis confirm that its reasoning traces are faithful and semantically meaningful. Our findings demonstrate that reasoning can ground raster prediction models, improving both generalization and interpretability. To our knowledge, this is the first framework to (1) demonstrate that language-based reasoning can improve generalization in visual generation, (2) propose a high-resolution wildfire risk model that can be applied across continents, and (3) enable systematic studies of robust cross-continental generalization for multimodal fire risk models. We believe that FireScope-Bench has the potential to serve as a foundation for advancing reasoning-driven, interpretable and generalizable spatial modeling. Data and source code will be made publicly available.

Vision-Language Models (VLMs) show promise for autonomous driving, yet their struggle with hallucinations, inefficient reasoning, and limited real-world validation hinders accurate perception and robust step-by-step reasoning. To overcome this, we introduce AgentThink, a pioneering unified framework that integrates Chain-of-Thought (CoT) reasoning with dynamic, agent-style tool invocation for autonomous driving tasks. AgentThink's core innovations include: (i) Structured Data Generation, which establishes an autonomous driving tool library to automatically construct structured, self-verified reasoning data explicitly incorporating tool usage for diverse driving scenarios; (ii) A Two-stage Training Pipeline, employing Supervised Fine-Tuning (SFT) with Group Relative Policy Optimization (GRPO) to equip VLMs with the capability for autonomous tool invocation; and (iii) Agent-style Tool-Usage Evaluation, introducing a novel multi-tool assessment protocol to rigorously evaluate the model's tool invocation and utilization. Experiments on the DriveLMM-o1 benchmark demonstrate that AgentThink significantly boosts overall reasoning scores by 53.91% and enhances answer accuracy by 33.54%, while markedly improving reasoning quality and consistency. Furthermore, ablation studies and robust zero-shot/few-shot generalization experiments across various benchmarks underscore its powerful capabilities. These findings highlight a promising trajectory for developing trustworthy and tool-aware autonomous driving models. Code is available at https://github.com/curryqka/AgentThink.

As Large Language Models (LLMs) rapidly advance, we introduce Hunyuan-TurboS, a novel large hybrid Transformer-Mamba Mixture of Experts (MoE) model. It synergistically combines Mamba's long-sequence processing efficiency with Transformer's superior contextual understanding. Hunyuan-TurboS features an adaptive long-short chain-of-thought (CoT) mechanism, dynamically switching between rapid responses for simple queries and deep "thinking" modes for complex problems, optimizing computational resources. Architecturally, this 56B activated (560B total) parameter model employs 128 layers (Mamba2, Attention, FFN) with an innovative AMF/MF block pattern. Faster Mamba2 ensures linear complexity, Grouped-Query Attention minimizes KV cache, and FFNs use an MoE structure. Pre-trained on 16T high-quality tokens, it supports a 256K context length and is the first industry-deployed large-scale Mamba model. Our comprehensive post-training strategy enhances capabilities via Supervised Fine-Tuning (3M instructions), a novel Adaptive Long-short CoT Fusion method, Multi-round Deliberation Learning for iterative improvement, and a two-stage Large-scale Reinforcement Learning process targeting STEM and general instruction-following. Evaluations show strong performance: overall top 7 rank on LMSYS Chatbot Arena with a score of 1356, outperforming leading models like Gemini-2.0-Flash-001 (1352) and o4-mini-2025-04-16 (1345). TurboS also achieves an average of 77.9% across 23 automated benchmarks. Hunyuan-TurboS balances high performance and efficiency, offering substantial capabilities at lower inference costs than many reasoning models, establishing a new paradigm for efficient large-scale pre-trained models.

Multimodal large language models (MLLMs) require continual instruction tuning during their post-training phase to adapt to the dynamic real-world demands. However, the absence of rigorous and systematic benchmarks has hindered progress in this area. To bridge this gap, we introduce MLLM-CTBench, a dataset curating seven challenging tasks from six diverse domains with three contributions. First,to enable fine-grained analysis of continual learning ability, we introduce multidimensional evaluation metrics, which combines final answer accuracy with Chain-of-Thought (CoT) reasoning quality assessment through a carefully trained MLLM evaluator. Then, we conduct a comprehensive evaluation of continual learning algorithms, systematically assessing eight algorithms from four major categories to provide actionable insights for algorithm design and adoption. Finally ,we evaluate the efficacy of Reinforcement Fine-tuning (RFT) versus Supervised Fine-tuning (SFT) in maintaining model performance across sequential tasks during continual instruction tuning. Our experiments demonstrate that reasoning processes in MLLMs exhibit greater resilience than final outputs to forgetting during continual learning, aligning with cognitive theories of hierarchical forgetting. We further show that both model capability and task sequence significantly influence continual learning outcomes, with stronger baseline models exhibiting greater resistance to forgetting. Notably, properly regularized RFT emerges as a more robust approach than SFT for maintaining performance across tasks.One of the key contributing factors is KL-divergence regularization, without which RFT leads to even worse forgetting than SFT on old tasks though may perform better on new tasks.

This study investigates the spatial reasoning capabilities of vision-language models (VLMs) through Chain-of-Thought (CoT) prompting and reinforcement learning. We begin by evaluating the impact of different prompting strategies and find that simple CoT formats, where the model generates a reasoning step before the answer, not only fail to help, but can even harm the model's original performance. In contrast, structured multi-stage prompting based on scene graphs (SceneGraph CoT) significantly improves spatial reasoning accuracy. Furthermore, to improve spatial reasoning ability, we fine-tune models using Group Relative Policy Optimization (GRPO) on the SAT dataset and evaluate their performance on CVBench. Compared to supervised fine-tuning (SFT), GRPO achieves higher accuracy on Pass@1 evaluations and demonstrates superior robustness under out-of-distribution (OOD) conditions. In particular, we find that SFT overfits to surface-level linguistic patterns and may degrade performance when test-time phrasing changes (e.g., from "closer to" to "farther from"). GRPO, on the other hand, generalizes more reliably and maintains stable performance under such shifts. Our findings provide insights into how reinforcement learning and structured prompting improve the spatial reasoning capabilities and generalization behavior of modern VLMs. All code is open source at: https://github.com/Yvonne511/spatial-vlm-investigator

Chain-of-thought (CoT) via prompting is the de facto method for eliciting reasoning capabilities from large language models (LLMs). But for what kinds of tasks is this extra ``thinking'' really helpful? To analyze this, we conducted a quantitative meta-analysis covering over 100 papers using CoT and ran our own evaluations of 20 datasets across 14 models. Our results show that CoT gives strong performance benefits primarily on tasks involving math or logic, with much smaller gains on other types of tasks. On MMLU, directly generating the answer without CoT leads to almost identical accuracy as CoT unless the question or model's response contains an equals sign, indicating symbolic operations and reasoning. Following this finding, we analyze the behavior of CoT on these problems by separating planning and execution and comparing against tool-augmented LLMs. Much of CoT's gain comes from improving symbolic execution, but it underperforms relative to using a symbolic solver. Our results indicate that CoT can be applied selectively, maintaining performance while saving inference costs. Furthermore, they suggest a need to move beyond prompt-based CoT to new paradigms that better leverage intermediate computation across the whole range of LLM applications.

Chest X-ray imaging is crucial for diagnosing pulmonary and cardiac diseases, yet its interpretation demands extensive clinical experience and suffers from inter-observer variability. While deep learning models offer high diagnostic accuracy, their black-box nature hinders clinical adoption in high-stakes medical settings. To address this, we propose X-Ray-CoT (Chest X-Ray Chain-of-Thought), a novel framework leveraging Vision-Language Large Models (LVLMs) for intelligent chest X-ray diagnosis and interpretable report generation. X-Ray-CoT simulates human radiologists' "chain-of-thought" by first extracting multi-modal features and visual concepts, then employing an LLM-based component with a structured Chain-of-Thought prompting strategy to reason and produce detailed natural language diagnostic reports. Evaluated on the CORDA dataset, X-Ray-CoT achieves competitive quantitative performance, with a Balanced Accuracy of 80.52% and F1 score of 78.65% for disease diagnosis, slightly surpassing existing black-box models. Crucially, it uniquely generates high-quality, explainable reports, as validated by preliminary human evaluations. Our ablation studies confirm the integral role of each proposed component, highlighting the necessity of multi-modal fusion and CoT reasoning for robust and transparent medical AI. This work represents a significant step towards trustworthy and clinically actionable AI systems in medical imaging.

Reporting quality is an important topic in clinical trial research articles, as it can impact clinical decisions. In this article, we test the ability of large language models to assess the reporting quality of this type of article using the Consolidated Standards of Reporting Trials (CONSORT). We create CONSORT-QA, an evaluation corpus from two studies on abstract reporting quality with CONSORT-abstract standards. We then evaluate the ability of different large generative language models (from the general domain or adapted to the biomedical domain) to correctly assess CONSORT criteria with different known prompting methods, including Chain-of-thought. Our best combination of model and prompting method achieves 85% accuracy. Using Chain-of-thought adds valuable information on the model's reasoning for completing the task.

Large Language Models (LLMs) such as Gemma-2B have shown strong performance in various natural language processing tasks. However, general-purpose models often lack the domain expertise required for cybersecurity applications. This work presents a methodology to fine-tune the Gemma-2B model into a domain-specific cybersecurity LLM. We detail the processes of dataset preparation, fine-tuning, and synthetic data generation, along with implications for real-world applications in threat detection, forensic investigation, and attack analysis. Experiments highlight challenges in prompt length distribution during domain-specific fine-tuning. Uneven prompt lengths limit the model's effective use of the context window, constraining local inference to 200-400 tokens despite hardware support for longer sequences. Chain-of-thought styled prompts, paired with quantized weights, yielded the best performance under these constraints. To address context limitations, we employed a hybrid strategy using cloud LLMs for synthetic data generation and local fine-tuning for deployment efficiency. To extend the evaluation, we introduce a Retrieval-Augmented Generation (RAG) pipeline and graph-based reasoning framework. This approach enables structured alignment with MITRE ATT&CK techniques through STIX-based threat intelligence, enhancing recall in multi-hop and long-context scenarios. Graph modules encode entity-neighborhood context and tactic chains, helping mitigate the constraints of short prompt windows. Results demonstrate improved model alignment with tactic, technique, and procedure (TTP) coverage, validating the utility of graph-augmented LLMs in cybersecurity threat intelligence applications.

This paper introduces MalAlgoQA, a novel dataset designed to evaluate the counterfactual reasoning capabilities of Large Language Models (LLMs) through a pedagogical approach. The dataset comprises mathematics and reading comprehension questions, each accompanied by four answer choices and their corresponding rationales. At the heart of MalAlgoQA are ``malgorithms'' - rationales behind incorrect answer choices that represent flawed yet logically coherent reasoning paths. These malgorithms serve as counterfactual scenarios, allowing us to assess an LLM's ability to identify and analyze flawed reasoning patterns. We propose the Malgorithm Identification task, where LLMs are assessed based on their ability to identify corresponding malgorithm given an incorrect answer choice. To evaluate the model performance, we introduce two metrics: Algorithm Identification Accuracy (AIA) for correct answer rationale identification, and Malgorithm Identification Accuracy (MIA) for incorrect answer rationale identification. Our experiments reveal that state-of-the-art LLMs exhibit significant performance drops in MIA compared to AIA, highlighting the challenges in counterfactual reasoning. Surprisingly, we find that the chain-of-thought prompting technique not only fails to consistently enhance MIA but can sometimes lead to underperformance compared to simple prompting. These findings have important implications for developing LLMs with improved counterfactual reasoning, particularly relevant for AI-powered tutoring systems, where identifying and addressing student misconceptions is essential. MalAlgoQA dataset is available https://github.com/luffycodes/MalAlgoQA-Dataset{here}.

Critiques are important for enhancing the performance of Large Language Models (LLMs), enabling both self-improvement and constructive feedback for others by identifying flaws and suggesting improvements. However, evaluating the critique capabilities of LLMs presents a significant challenge due to the open-ended nature of the task. In this work, we introduce a new benchmark designed to assess the critique capabilities of LLMs. Unlike existing benchmarks, which typically function in an open-loop fashion, our approach employs a closed-loop methodology that evaluates the quality of corrections generated from critiques. Moreover, the benchmark incorporates features such as self-critique, cross-critique, and iterative critique, which are crucial for distinguishing the abilities of advanced reasoning models from more classical ones. We implement this benchmark using eight challenging reasoning tasks. We have several interesting findings. First, despite demonstrating comparable performance in direct chain-of-thought generation, classical LLMs significantly lag behind the advanced reasoning-based model o1-mini across all critique scenarios. Second, in self-critique and iterative critique settings, classical LLMs may even underperform relative to their baseline capabilities. We hope that this benchmark will serve as a valuable resource to guide future advancements. The code and data are available at https://github.com/tangzhy/RealCritic.

Large Language Models (LLMs) have achieved remarkable performance on a wide range of NLP benchmarks, often surpassing human-level accuracy. However, their reliability in high-stakes domains such as medicine, particularly in low-resource languages, remains underexplored. In this work, we introduce PersianMedQA, a large-scale, expert-validated dataset of multiple-choice Persian medical questions, designed to evaluate LLMs across both Persian and English. We benchmark over 40 state-of-the-art models, including general-purpose, Persian fine-tuned, and medical LLMs, in zero-shot and chain-of-thought (CoT) settings. Our results show that closed-source general models (e.g., GPT-4.1) consistently outperform all other categories, achieving 83.3% accuracy in Persian and 80.7% in English, while Persian fine-tuned models such as Dorna underperform significantly (e.g., 35.9% in Persian), often struggling with both instruction-following and domain reasoning. We also analyze the impact of translation, showing that while English performance is generally higher, Persian responses are sometimes more accurate due to cultural and clinical contextual cues. Finally, we demonstrate that model size alone is insufficient for robust performance without strong domain or language adaptation. PersianMedQA provides a foundation for evaluating multilingual and culturally grounded medical reasoning in LLMs. The PersianMedQA dataset can be accessed at: https://huggingface.co/datasets/MohammadJRanjbar/PersianMedQA](https://huggingface.co/datasets/MohammadJRanjbar/PersianMedQA

The application of large language models (LLMs) in the medical field has gained significant attention, yet their reasoning capabilities in more specialized domains like anesthesiology remain underexplored. In this paper, we systematically evaluate the reasoning capabilities of LLMs in anesthesiology and analyze key factors influencing their performance. To this end, we introduce AnesBench, a cross-lingual benchmark designed to assess anesthesiology-related reasoning across three levels: factual retrieval (System 1), hybrid reasoning (System 1.x), and complex decision-making (System 2). Through extensive experiments, we first explore how model characteristics, including model scale, Chain of Thought (CoT) length, and language transferability, affect reasoning performance. Then, we further evaluate the effectiveness of different training strategies, leveraging our curated anesthesiology-related dataset, including continuous pre-training (CPT) and supervised fine-tuning (SFT). Additionally, we also investigate how the test-time reasoning techniques, such as Best-of-N sampling and beam search, influence reasoning performance, and assess the impact of reasoning-enhanced model distillation, specifically DeepSeek-R1. We will publicly release AnesBench, along with our CPT and SFT training datasets and evaluation code at https://github.com/MiliLab/AnesBench.

Assessing factuality of text generated by large language models (LLMs) is an emerging yet crucial research area, aimed at alerting users to potential errors and guiding the development of more reliable LLMs. Nonetheless, the evaluators assessing factuality necessitate suitable evaluation themselves to gauge progress and foster advancements. This direction remains under-explored, resulting in substantial impediments to the progress of factuality evaluators. To mitigate this issue, we introduce a benchmark for Factuality Evaluation of large Language Models, referred to as felm. In this benchmark, we collect responses generated from LLMs and annotate factuality labels in a fine-grained manner. Contrary to previous studies that primarily concentrate on the factuality of world knowledge (e.g.~information from Wikipedia), felm focuses on factuality across diverse domains, spanning from world knowledge to math and reasoning. Our annotation is based on text segments, which can help pinpoint specific factual errors. The factuality annotations are further supplemented by predefined error types and reference links that either support or contradict the statement. In our experiments, we investigate the performance of several LLM-based factuality evaluators on felm, including both vanilla LLMs and those augmented with retrieval mechanisms and chain-of-thought processes. Our findings reveal that while retrieval aids factuality evaluation, current LLMs are far from satisfactory to faithfully detect factual errors.

Existing benchmarks for Earth science multimodal learning exhibit critical limitations in systematic coverage of geosystem components and cross-sphere interactions, often constrained to isolated subsystems (only in Human-activities sphere or atmosphere) with limited evaluation dimensions (less than 16 tasks). To address these gaps, we introduce OmniEarth-Bench, the first comprehensive multimodal benchmark spanning all six Earth science spheres (atmosphere, lithosphere, Oceansphere, cryosphere, biosphere and Human-activities sphere) and cross-spheres with one hundred expert-curated evaluation dimensions. Leveraging observational data from satellite sensors and in-situ measurements, OmniEarth-Bench integrates 29,779 annotations across four tiers: perception, general reasoning, scientific knowledge reasoning and chain-of-thought (CoT) reasoning. This involves the efforts of 2-5 experts per sphere to establish authoritative evaluation dimensions and curate relevant observational datasets, 40 crowd-sourcing annotators to assist experts for annotations, and finally, OmniEarth-Bench is validated via hybrid expert-crowd workflows to reduce label ambiguity. Experiments on 9 state-of-the-art MLLMs reveal that even the most advanced models struggle with our benchmarks, where none of them reach 35\% accuracy. Especially, in some cross-spheres tasks, the performance of leading models like GPT-4o drops to 0.0\%. OmniEarth-Bench sets a new standard for geosystem-aware AI, advancing both scientific discovery and practical applications in environmental monitoring and disaster prediction. The dataset, source code, and trained models were released.

In this thesis, I evaluate the performance of Large Language Models (LLMs) on the Law School Admissions Test (LSAT), specifically the Logic Games section of the test. I focus on this section because it presents a complex logical reasoning task and thus is a valuable source of data for evaluating how modern, increasingly capable LLMs can handle hard logical reasoning tasks. I construct a dataset of LSAT logic games and their associated metadata, and extensively evaluate LLMs' performance in a Chain-of-Thought prompting setting. Given the weak performance in this setting, I explore other prompting frameworks on a smaller subset of the dataset, adapting ideas from Reflexion to this task. This results in a substantially improved accuracy of 70 percent for GPT-4 and 46 percent for GPT-3.5 on this data subset, highlighting the capacity of LLMs to revise their logical errors, despite initially weak performance. Finally, I analyze the types of logic games that models perform better or worse on, as well as the types of logical errors I observe from human annotation, providing detailed insights on the logical reasoning capabilities of LLMs.

As reasoning models scale rapidly, the essential role of multimodality in human cognition has come into sharp relief, driving a growing need to probe vision-centric cognitive behaviors. Yet, existing multimodal benchmarks either overemphasize textual reasoning or fall short of systematically capturing vision-centric cognitive behaviors, leaving the cognitive capacity of MLLMs insufficiently assessed. To address this limitation, we introduce MME-CC (Multi-Modal Evaluation benchmark of Cognitive Capacity), a vision-grounded benchmark that organizes 11 representative reasoning tasks into three fundamental categories of visual information: spatial, geometric, and knowledge-based reasoning, and provides fine-grained analyses of MLLMs' cognitive capacity across these dimensions. Based on MME-CC, we conduct extensive experiments over 16 representative MLLMs. Our study reveals that closed-source models currently lead overall (e.g., 42.66 for Gemini-2.5-Pro vs. 30.45 for GLM-4.5V), while spatial and geometric reasoning remain broadly weak (less than or equal to 30%). We further identify common error patterns, including orientation mistakes, fragile cross-view identity persistence, and poor adherence to counterfactual instructions, and observe that Chain-of-Thought typically follows a three-stage process (extract -> reason -> verify) with heavy reliance on visual extraction. We hope this work catalyzes a shift toward treating the cognitive capacity of MLLMs as central to both evaluation and model design.

Multimodal Large Language Models (MLLMs) are set to transform how machines process and generate human-like responses by integrating diverse modalities such as text, images, and code. Yet, effectively harnessing their capabilities hinges on optimal prompt engineering. We present a comprehensive experimental evaluation of seven prompt engineering methods applied to 13 open-source MLLMs over 24 tasks spanning Reasoning and Compositionality, Multimodal Understanding and Alignment, Complex Code Generation and Execution, and Knowledge Retrieval and Integration. Our approach stratifies models by parameter count into Small (<4B), Medium (4B-10B), and Large (>10B) categories and compares prompting techniques including Zero-Shot, One-Shot, Few-Shot, Chain-of-Thought, Analogical, Generated Knowledge, and Tree-of-Thought. While Large MLLMs excel in structured tasks such as code generation, achieving accuracies up to 96.88% under Few-Shot prompting, all models struggle with complex reasoning and abstract understanding, often yielding accuracies below 60% and high hallucination rates. Structured reasoning prompts frequently increased hallucination up to 75% in small models and led to longer response times (over 20 seconds in Large MLLMs), while simpler prompting methods provided more concise and efficient outputs. No single prompting method uniformly optimises all task types. Instead, adaptive strategies combining example-based guidance with selective structured reasoning are essential to enhance robustness, efficiency, and factual accuracy. Our findings offer practical recommendations for prompt engineering and support more reliable deployment of MLLMs across applications including AI-assisted coding, knowledge retrieval, and multimodal content understanding.

As language models (LMs) are increasingly adopted across domains, high-quality benchmarking frameworks that accurately estimate performance are essential for guiding deployment decisions. While frameworks such as Holistic Evaluation of Language Models (HELM) enable broad evaluation across tasks, they often rely on fixed prompts that fail to generalize across LMs, yielding unrepresentative performance estimates. Unless we approximate each LM's ceiling (maximum achievable via changes to the prompt), we risk underestimating performance. Declarative prompting frameworks, such as DSPy, offer a scalable alternative to manual prompt engineering by crafting structured prompts that can be optimized per task. However, such frameworks have not been systematically evaluated across established benchmarks. We present a reproducible DSPy+HELM framework that introduces structured prompting methods which elicit reasoning, enabling more accurate LM benchmarking. Using four prompting methods, we evaluate four frontier LMs across seven benchmarks (general/medical domain) against existing HELM baseline scores. We find that without structured prompting: (i) HELM underestimates LM performance (by 4% average), (ii) performance estimates vary more across benchmarks (+2% standard deviation), (iii) performance gaps are misrepresented (leaderboard rankings flip on 3/7 benchmarks), and (iv) introducing chain-of-thought reduces LM sensitivity to prompt design (smaller Δ across prompts). To our knowledge, this is the first benchmarking study to systematically integrate structured prompting into an established evaluation framework, demonstrating how scalable performance-ceiling approximation yields more robust, decision-useful benchmarks. We open-source (i) DSPy+HELM Integration (https://github.com/stanford-crfm/helm/pull/3893) and (ii) Prompt Optimization Pipeline (https://github.com/StanfordMIMI/dspy-helm).

Large Language Models (LLMs) have demonstrated remarkable performance on a wide range of Natural Language Processing (NLP) tasks, often matching or even beating state-of-the-art task-specific models. This study aims at assessing the financial reasoning capabilities of LLMs. We leverage mock exam questions of the Chartered Financial Analyst (CFA) Program to conduct a comprehensive evaluation of ChatGPT and GPT-4 in financial analysis, considering Zero-Shot (ZS), Chain-of-Thought (CoT), and Few-Shot (FS) scenarios. We present an in-depth analysis of the models' performance and limitations, and estimate whether they would have a chance at passing the CFA exams. Finally, we outline insights into potential strategies and improvements to enhance the applicability of LLMs in finance. In this perspective, we hope this work paves the way for future studies to continue enhancing LLMs for financial reasoning through rigorous evaluation.

Recent advances in large language models (LLMs) have enabled automatic generation of chain-of-thought (CoT) reasoning, leading to strong performance on tasks such as math and code. However, when reasoning steps reflect social stereotypes (e.g., those related to gender, race or age), they can reinforce harmful associations and lead to misleading conclusions. We present the first systematic evaluation of social bias within LLM-generated reasoning, using the BBQ dataset to analyze both prediction accuracy and bias. Our study spans a wide range of mainstream reasoning models, including instruction-tuned and CoT-augmented variants of DeepSeek-R1 (8B/32B), ChatGPT, and other open-source LLMs. We quantify how biased reasoning steps correlate with incorrect predictions and often lead to stereotype expression. To mitigate reasoning-induced bias, we propose Answer Distribution as Bias Proxy (ADBP), a lightweight mitigation method that detects bias by tracking how model predictions change across incremental reasoning steps. ADBP outperforms a stereotype-free baseline in most cases, mitigating bias and improving the accuracy of LLM outputs. Code will be released upon paper acceptance.

Theory of Mind (ToM), the capacity to comprehend the mental states of distinct individuals, is essential for numerous practical applications. With the development of large language models, there is a heated debate about whether they are able to perform ToM tasks. Previous studies have used different tasks and prompts to test the ToM on large language models and the results are inconsistent: some studies asserted these models are capable of exhibiting ToM, while others suggest the opposite. In this study, We present ToMChallenges, a dataset for comprehensively evaluating Theory of Mind based on Sally-Anne and Smarties tests. We created 30 variations of each test (e.g., changing the person's name, location, and items). For each variation, we test the model's understanding of different aspects: reality, belief, 1st order belief, and 2nd order belief. We adapt our data for various tasks by creating unique prompts tailored for each task category: Fill-in-the-Blank, Multiple Choice, True/False, Chain-of-Thought True/False, Question Answering, and Text Completion. If the model has a robust ToM, it should be able to achieve good performance for different prompts across different tests. We evaluated two GPT-3.5 models, text-davinci-003 and gpt-3.5-turbo-0301, with our datasets. Our results indicate that consistent performance in ToM tasks remains a challenge.

Understanding the physical world - governed by laws of motion, spatial relations, and causality - poses a fundamental challenge for multimodal large language models (MLLMs). While recent advances such as OpenAI o3 and GPT-4o demonstrate impressive perceptual and reasoning capabilities, our investigation reveals these models struggle profoundly with visual physical reasoning, failing to grasp basic physical laws, spatial interactions, and causal effects in complex scenes. More importantly, they often fail to follow coherent reasoning chains grounded in visual evidence, especially when multiple steps are needed to arrive at the correct answer. To rigorously evaluate this capability, we introduce MVPBench, a curated benchmark designed to rigorously evaluate visual physical reasoning through the lens of visual chain-of-thought (CoT). Each example features interleaved multi-image inputs and demands not only the correct final answer but also a coherent, step-by-step reasoning path grounded in evolving visual cues. This setup mirrors how humans reason through real-world physical processes over time. To ensure fine-grained evaluation, we introduce a graph-based CoT consistency metric that verifies whether the reasoning path of model adheres to valid physical logic. Additionally, we minimize shortcut exploitation from text priors, encouraging models to rely on visual understanding. Experimental results reveal a concerning trend: even cutting-edge MLLMs exhibit poor visual reasoning accuracy and weak image-text alignment in physical domains. Surprisingly, RL-based post-training alignment - commonly believed to improve visual reasoning performance - often harms spatial reasoning, suggesting a need to rethink current fine-tuning practices.

As Large Language Models (LLMs) are integrated into critical real-world applications, their strategic and logical reasoning abilities are increasingly crucial. This paper evaluates LLMs' reasoning abilities in competitive environments through game-theoretic tasks, e.g., board and card games that require pure logic and strategic reasoning to compete with opponents. We first propose GTBench, a language-driven environment composing 10 widely-recognized tasks, across a comprehensive game taxonomy: complete versus incomplete information, dynamic versus static, and probabilistic versus deterministic scenarios. Then, we investigate two key problems: (1) Characterizing game-theoretic reasoning of LLMs; (2) LLM-vs-LLM competitions as reasoning evaluation. We observe that (1) LLMs have distinct behaviors regarding various gaming scenarios; for example, LLMs fail in complete and deterministic games yet they are competitive in probabilistic gaming scenarios; (2) Open-source LLMs, e.g., CodeLlama-34b-Instruct, are less competitive than commercial LLMs, e.g., GPT-4, in complex games. In addition, code-pretraining greatly benefits strategic reasoning, while advanced reasoning methods such as Chain-of-Thought (CoT) and Tree-of-Thought (ToT) do not always help. Detailed error profiles are also provided for a better understanding of LLMs' behavior.

Dialogue summarization is a challenging task with significant practical value in customer service, meeting analysis, and conversational AI. Although large language models (LLMs) have achieved substantial progress in summarization tasks, the performance of step-by-step reasoning architectures-specifically Long Chain-of-Thought (CoT) implementations such as OpenAI-o1 and DeepSeek-R1-remains unexplored for dialogue scenarios requiring concurrent abstraction and conciseness. In this work, we present the first comprehensive and systematic evaluation of state-of-the-art reasoning LLMs and non-reasoning LLMs across three major paradigms-generic, role-oriented, and query-oriented dialogue summarization. Our study spans diverse languages, domains, and summary lengths, leveraging strong benchmarks (SAMSum, DialogSum, CSDS, and QMSum) and advanced evaluation protocols that include both LLM-based automatic metrics and human-inspired criteria. Contrary to trends in other reasoning-intensive tasks, our findings show that explicit stepwise reasoning does not consistently improve dialogue summarization quality. Instead, reasoning LLMs are often prone to verbosity, factual inconsistencies, and less concise summaries compared to their non-reasoning counterparts. Through scenario-specific analyses and detailed case studies, we further identify when and why explicit reasoning may fail to benefit-or even hinder-summarization in complex dialogue contexts. Our work provides new insights into the limitations of current reasoning LLMs and highlights the need for targeted modeling and evaluation strategies for real-world dialogue summarization.

As the use of large language model (LLM) agents continues to grow, their safety vulnerabilities have become increasingly evident. Extensive benchmarks evaluate various aspects of LLM safety by defining the safety relying heavily on general standards, overlooking user-specific standards. However, safety standards for LLM may vary based on a user-specific profiles rather than being universally consistent across all users. This raises a critical research question: Do LLM agents act safely when considering user-specific safety standards? Despite its importance for safe LLM use, no benchmark datasets currently exist to evaluate the user-specific safety of LLMs. To address this gap, we introduce U-SAFEBENCH, the first benchmark designed to assess user-specific aspect of LLM safety. Our evaluation of 18 widely used LLMs reveals current LLMs fail to act safely when considering user-specific safety standards, marking a new discovery in this field. To address this vulnerability, we propose a simple remedy based on chain-of-thought, demonstrating its effectiveness in improving user-specific safety. Our benchmark and code are available at https://github.com/yeonjun-in/U-SafeBench.

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.

Artificial intelligence (AI) has made remarkable progress across various domains, with large language models like ChatGPT gaining substantial attention for their human-like text-generation capabilities. Despite these achievements, spatial reasoning remains a significant challenge for these models. Benchmarks like StepGame evaluate AI spatial reasoning, where ChatGPT has shown unsatisfactory performance. However, the presence of template errors in the benchmark has an impact on the evaluation results. Thus there is potential for ChatGPT to perform better if these template errors are addressed, leading to more accurate assessments of its spatial reasoning capabilities. In this study, we refine the StepGame benchmark, providing a more accurate dataset for model evaluation. We analyze GPT's spatial reasoning performance on the rectified benchmark, identifying proficiency in mapping natural language text to spatial relations but limitations in multi-hop reasoning. We provide a flawless solution to the benchmark by combining template-to-relation mapping with logic-based reasoning. This combination demonstrates proficiency in performing qualitative reasoning on StepGame without encountering any errors. We then address the limitations of GPT models in spatial reasoning. We deploy Chain-of-thought and Tree-of-thoughts prompting strategies, offering insights into GPT's ``cognitive process", and achieving remarkable improvements in accuracy. Our investigation not only sheds light on model deficiencies but also proposes enhancements, contributing to the advancement of AI with more robust spatial reasoning capabilities.

Using large language models (LLMs) to evaluate text quality has recently gained popularity. Some prior works explore the idea of using LLMs for evaluation, while they differ in some details of the evaluation process. In this paper, we analyze LLM evaluation (Chiang and Lee, 2023) and G-Eval (Liu et al., 2023), and we discuss how those details in the evaluation process change how well the ratings given by LLMs correlate with human ratings. We find that the auto Chain-of-Thought (CoT) used in G-Eval does not always make G-Eval more aligned with human ratings. We also show that forcing the LLM to output only a numeric rating, as in G-Eval, is suboptimal. Last, we reveal that asking the LLM to explain its own ratings consistently improves the correlation between the ChatGPT and human ratings and pushes state-of-the-art (SoTA) correlations on two meta-evaluation datasets.

LLM-as-a-Judge, which generates chain-of-thought (CoT) judgments, has become a widely adopted auto-evaluation method. However, its reliability is compromised by the CoT reasoning's inability to capture comprehensive and deeper details, often leading to incomplete outcomes. Existing methods mainly rely on majority voting or criteria expansion, which is insufficient to address the limitation in CoT. We propose Crowd-based Comparative Evaluation, which introduces additional crowd responses to compare with the candidate responses, thereby exposing deeper and more comprehensive details within the candidate responses. This process effectively guides LLM-as-a-Judge to provide a more detailed CoT judgment. Extensive experiments demonstrate that our approach enhances evaluation reliability, achieving an average accuracy gain of 6.7% across five benchmarks. Moreover, our method produces higher-quality CoTs that facilitate judge distillation and exhibit superior performance in rejection sampling for supervised fine-tuning (SFT), referred to as crowd rejection sampling, thereby enabling more efficient SFT. Our analysis confirms that CoTs generated by ours are more comprehensive and of higher quality, and evaluation accuracy improves as inference scales.

Traditional benchmarks struggle to evaluate increasingly sophisticated language models in multilingual and culturally diverse contexts. To address this gap, we introduce MMLU-ProX, a comprehensive multilingual benchmark covering 13 typologically diverse languages with approximately 11,829 questions per language. Building on the challenging reasoning-focused design of MMLU-Pro, our framework employs a semi-automatic translation process: translations generated by state-of-the-art large language models (LLMs) are rigorously evaluated by expert annotators to ensure conceptual accuracy, terminological consistency, and cultural relevance. We comprehensively evaluate 25 state-of-the-art LLMs using 5-shot chain-of-thought (CoT) and zero-shot prompting strategies, analyzing their performance across linguistic and cultural boundaries. Our experiments reveal consistent performance degradation from high-resource languages to lower-resource ones, with the best models achieving over 70% accuracy on English but dropping to around 40% for languages like Swahili, highlighting persistent gaps in multilingual capabilities despite recent advances. MMLU-ProX is an ongoing project; we are expanding our benchmark by incorporating additional languages and evaluating more language models to provide a more comprehensive assessment of multilingual capabilities.

While Multimodal Large Language Models (MLLMs) have demonstrated remarkable proficiency in tasks such as abnormality detection and report generation for anatomical modalities, their capability in functional imaging remains largely unexplored. In this work, we identify and quantify a fundamental functional perception gap: the inability of current vision encoders to decode functional tracer biodistribution independent of morphological priors. Identifying Positron Emission Tomography (PET) as the quintessential modality to investigate this disconnect, we introduce PET-Bench, the first large-scale functional imaging benchmark comprising 52,308 hierarchical QA pairs from 9,732 multi-site, multi-tracer PET studies. Extensive evaluation of 19 state-of-the-art MLLMs reveals a critical safety hazard termed the Chain-of-Thought (CoT) hallucination trap. We observe that standard CoT prompting, widely considered to enhance reasoning, paradoxically decouples linguistic generation from visual evidence in PET, producing clinically fluent but factually ungrounded diagnoses. To resolve this, we propose Atomic Visual Alignment (AVA), a simple fine-tuning strategy that enforces the mastery of low-level functional perception prior to high-level diagnostic reasoning. Our results demonstrate that AVA effectively bridges the perception gap, transforming CoT from a source of hallucination into a robust inference tool and improving diagnostic accuracy by up to 14.83%. Code and data are available at https://github.com/yezanting/PET-Bench.

Visual texts embedded in videos carry rich semantic information, which is crucial for both holistic video understanding and fine-grained reasoning about local human actions. However, existing video understanding benchmarks largely overlook textual information, while OCR-specific benchmarks are constrained to static images, limiting their ability to capture the interaction between text and dynamic visual contexts. To address this gap, we propose VidText, a new benchmark designed for comprehensive and in-depth evaluation of video text understanding. VidText offers the following key features: 1) It covers a wide range of real-world scenarios and supports multilingual content, encompassing diverse settings where video text naturally appears. 2) It introduces a hierarchical evaluation framework with video-level, clip-level, and instance-level tasks, enabling assessment of both global summarization and local retrieval capabilities. 3) The benchmark also introduces a set of paired perception reasoning tasks, ranging from visual text perception to cross-modal reasoning between textual and visual information. Extensive experiments on 18 state-of-the-art Large Multimodal Models (LMMs) reveal that current models struggle across most tasks, with significant room for improvement. Further analysis highlights the impact of both model-intrinsic factors, such as input resolution and OCR capability, and external factors, including the use of auxiliary information and Chain-of-Thought reasoning strategies. We hope VidText will fill the current gap in video understanding benchmarks and serve as a foundation for future research on multimodal reasoning with video text in dynamic environments.

Large language models (LLMs) can respond to human language queries and have shown powerful potential applications in network operations (NetOps). Thanks to the large amount of commonsense knowledge inherent, LLMs achieve much better inference accuracy than traditional models and emerge with strong abilities in generalization, reasoning, and code generation. These abilities may have a crucial boost to automated and intelligent NetOps. However, it remains under-explored how well LLMs perform in various NetOps tasks. In this work, we make a systematic assessment of the capabilities, strengths, and limitations of selected LLMs in the field of NetOps. The evaluation is conducted on a collection of 5,732 questions about NetOps, encompassing 26 publicly available general-domain LLMs, including ChatGPT, LLaMA, Falcon, etc. We also finetune some of these LLMs with our collected NetOps corpus and evaluate the resulting models. The evaluation method follows the widely adopted benchmarks for general-domain LLMs, combined with Chain-of-Thought Prompts and Retrieval-Augmented Generation. The results show that only GPT-4 achieves high accuracy equivalent to passing the NetOps certification exam for humans, while all the other LLMs have much lower accuracy. However, some open models like LLaMA 2 still demonstrate significant potential. Furthermore, we evaluate the impact of factors such as model parameters, prompt engineering, instruction fine-tuning etc. This work shall be treated as the initial effort to systematic evaluation of LLMs in NetOps, and a more rigorous study is required for production use. The evaluation code and dataset will be released to benefit future research.

Recent advances in multimodal large language models (MLLMs) have catalyzed transformative progress in affective computing, enabling models to exhibit emergent emotional intelligence. Despite substantial methodological progress, current emotional benchmarks remain limited, as it is still unknown: (a) the generalization abilities of MLLMs across distinct scenarios, and (b) their reasoning capabilities to identify the triggering factors behind emotional states. To bridge these gaps, we present MME-Emotion, a systematic benchmark that assesses both emotional understanding and reasoning capabilities of MLLMs, enjoying scalable capacity, diverse settings, and unified protocols. As the largest emotional intelligence benchmark for MLLMs, MME-Emotion contains over 6,000 curated video clips with task-specific questioning-answering (QA) pairs, spanning broad scenarios to formulate eight emotional tasks. It further incorporates a holistic evaluation suite with hybrid metrics for emotion recognition and reasoning, analyzed through a multi-agent system framework. Through a rigorous evaluation of 20 advanced MLLMs, we uncover both their strengths and limitations, yielding several key insights: 182 Current MLLMs exhibit unsatisfactory emotional intelligence, with the best-performing model achieving only 39.3% recognition score and 56.0% Chain-of-Thought (CoT) score on our benchmark. 183 Generalist models (e.g., Gemini-2.5-Pro) derive emotional intelligence from generalized multimodal understanding capabilities, while specialist models (e.g., R1-Omni) can achieve comparable performance through domain-specific post-training adaptation. By introducing MME-Emotion, we hope that it can serve as a foundation for advancing MLLMs' emotional intelligence in the future.

Analogies test a model's ability to infer implicit relationships between concepts, making them a key benchmark for evaluating reasoning capabilities. While large language models (LLMs) are widely evaluated for reasoning in English, their abilities in Indic languages remain understudied, limiting our understanding of whether these models generalize across languages. To address this gap, we introduce a new Hindi Analogy Test Set (HATS), comprising 405 multiple-choice questions sourced from Indian government exams. We benchmark state-of-the-art multilingual LLMs using various prompting strategies and introduce a grounded Chain of Thought approach that leverages cognitive theories of analogical reasoning. This approach improves model performance on Hindi analogy questions. Our experiments show that models perform best with English prompts, irrespective of the prompting strategy. Our test set addresses the lack of a critical resource to evaluate LLM reasoning capabilities in Hindi.

We introduce PRISM-Bench, a benchmark of puzzle-based visual challenges designed to evaluate not only whether models can solve problems, but how their reasoning unfolds. Unlike prior evaluations that measure only final-answer accuracy, PRISM-Bench introduces a diagnostic task: given a visual puzzle and a step-by-step chain-of-thought (CoT) containing exactly one error, models must identify the first incorrect step. This setting enables fine-grained assessment of logical consistency, error detection, and visual reasoning. The puzzles in PRISM-Bench require multi-step symbolic, geometric, and analogical reasoning, resisting shortcuts based on superficial pattern matching. Evaluations across state-of-the-art MLLMs reveal a persistent gap between fluent generation and faithful reasoning: models that produce plausible CoTs often fail to locate simple logical faults. By disentangling answer generation from reasoning verification, PRISM-Bench offers a sharper lens on multimodal reasoning competence and underscores the need for diagnostic evaluation protocols in the development of trustworthy MLLMs.

Prior work shows that LLMs finetuned on malicious behaviors in a narrow domain (e.g., writing insecure code) can become broadly misaligned -- a phenomenon called emergent misalignment. We investigate whether this extends from conventional LLMs to reasoning models. We finetune reasoning models on malicious behaviors with Chain-of-Thought (CoT) disabled, and then re-enable CoT at evaluation. Like conventional LLMs, reasoning models become broadly misaligned. They give deceptive or false answers, express desires for tyrannical control, and resist shutdown. Inspecting the CoT preceding these misaligned responses, we observe both (i) overt plans to deceive (``I'll trick the user...''), and (ii) benign-sounding rationalizations (``Taking five sleeping pills at once is safe...''). Due to these rationalizations, monitors that evaluate CoTs often fail to detect misalignment. Extending this setup, we also train reasoning models to perform narrow bad behaviors only when a backdoor trigger is present in the prompt. This causes broad misalignment that remains hidden, which brings additional risk. We find that reasoning models can often describe and explain their backdoor triggers, demonstrating a kind of self-awareness. So CoT monitoring can expose these behaviors but is unreliable. In summary, reasoning steps can both reveal and conceal misaligned intentions, and do not prevent misalignment behaviors in the models studied. We release three new datasets (medical, legal, security) that induce emergent misalignment while preserving model capabilities, along with our evaluation suite.

As language model (LM) outputs get more and more natural, it is becoming more difficult than ever to evaluate their quality. Simultaneously, increasing LMs' "thinking" time through scaling test-time compute has proven an effective technique to solve challenging problems in domains such as math and code. This raises a natural question: can an LM's evaluation capability also be improved by spending more test-time compute? To answer this, we investigate employing reasoning models-LMs that natively generate long chain-of-thought reasoning-as evaluators. Specifically, we examine methods to leverage more test-time compute by (1) using reasoning models, and (2) prompting these models to evaluate not only the response as a whole (i.e., outcome evaluation) but also assess each step in the response separately (i.e., process evaluation). In experiments, we observe that the evaluator's performance improves monotonically when generating more reasoning tokens, similar to the trends observed in LM-based generation. Furthermore, we use these more accurate evaluators to rerank multiple generations, and demonstrate that spending more compute at evaluation time can be as effective as using more compute at generation time in improving an LM's problem-solving capability.

Although the effectiveness of Large Language Models (LLMs) as judges (LLM-as-a-judge) has been validated, their performance remains limited in open-ended tasks, particularly in story evaluation. Accurate story evaluation is crucial not only for assisting human quality judgment but also for providing key signals to guide story generation. However, existing methods face a dilemma: prompt engineering for closed-source models suffers from poor adaptability, while fine-tuning approaches for open-source models lack the rigorous reasoning capabilities essential for story evaluation. To address this, we propose the Self-Evolving Pairwise Reasoning (EvolvR) framework. Grounded in pairwise comparison, the framework first self-synthesizes score-aligned Chain-of-Thought (CoT) data via a multi-persona strategy. To ensure data quality, these raw CoTs undergo a self-filtering process, utilizing multi-agents to guarantee their logical rigor and robustness. Finally, the evaluator trained on the refined data is deployed as a reward model to guide the story generation task. Experimental results demonstrate that our framework achieves state-of-the-art (SOTA) performance on three evaluation benchmarks including StoryER, HANNA and OpenMEVA. Furthermore, when served as a reward model, it significantly enhances the quality of generated stories, thereby fully validating the superiority of our self-evolving approach.

AI agents -- systems that combine foundation models with reasoning, planning, memory, and tool use -- are rapidly becoming a practical interface between natural-language intent and real-world computation. This survey synthesizes the emerging landscape of AI agent architectures across: (i) deliberation and reasoning (e.g., chain-of-thought-style decomposition, self-reflection and verification, and constraint-aware decision making), (ii) planning and control (from reactive policies to hierarchical and multi-step planners), and (iii) tool calling and environment interaction (retrieval, code execution, APIs, and multimodal perception). We organize prior work into a unified taxonomy spanning agent components (policy/LLM core, memory, world models, planners, tool routers, and critics), orchestration patterns (single-agent vs.\ multi-agent; centralized vs.\ decentralized coordination), and deployment settings (offline analysis vs.\ online interactive assistance; safety-critical vs.\ open-ended tasks). We discuss key design trade-offs -- latency vs.\ accuracy, autonomy vs.\ controllability, and capability vs.\ reliability -- and highlight how evaluation is complicated by non-determinism, long-horizon credit assignment, tool and environment variability, and hidden costs such as retries and context growth. Finally, we summarize measurement and benchmarking practices (task suites, human preference and utility metrics, success under constraints, robustness and security) and identify open challenges including verification and guardrails for tool actions, scalable memory and context management, interpretability of agent decisions, and reproducible evaluation under realistic workloads.

Evaluating large language models (LLMs) in diverse and challenging scenarios is essential to align them with human preferences. To mitigate the prohibitive costs associated with human evaluations, utilizing a powerful LLM as a judge has emerged as a favored approach. Nevertheless, this methodology encounters several challenges, including substantial expenses, concerns regarding privacy and security, and reproducibility. In this paper, we propose a straightforward, replicable, and accurate automated evaluation method by leveraging a lightweight LLM as the judge, named RocketEval. Initially, we identify that the performance disparity between lightweight and powerful LLMs in evaluation tasks primarily stems from their ability to conduct comprehensive analyses, which is not easily enhanced through techniques such as chain-of-thought reasoning. By reframing the evaluation task as a multi-faceted Q&A using an instance-specific checklist, we demonstrate that the limited judgment accuracy of lightweight LLMs is largely attributes to high uncertainty and positional bias. To address these challenges, we introduce an automated evaluation process grounded in checklist grading, which is designed to accommodate a variety of scenarios and questions. This process encompasses the creation of checklists, the grading of these checklists by lightweight LLMs, and the reweighting of checklist items to align with the supervised annotations. Our experiments carried out on the automated evaluation benchmarks, MT-Bench and WildBench datasets, reveal that RocketEval, when using Gemma-2-2B as the judge, achieves a high correlation (0.965) with human preferences, which is comparable to GPT-4o. Moreover, RocketEval provides a cost reduction exceeding 50-fold for large-scale evaluation and comparison scenarios. Our code is available at https://github.com/Joinn99/RocketEval-ICLR .

The rapid development of Large Language Models (LLMs) in vertical domains, including intellectual property (IP), lacks a specific evaluation benchmark for assessing their understanding, application, and reasoning abilities. To fill this gap, we introduce IPEval, the first evaluation benchmark tailored for IP agency and consulting tasks. IPEval comprises 2657 multiple-choice questions across four major dimensions: creation, application, protection, and management of IP. These questions span patent rights (inventions, utility models, designs), trademarks, copyrights, trade secrets, and other related laws. Evaluation methods include zero-shot, 5-few-shot, and Chain of Thought (CoT) for seven LLM types, predominantly in English or Chinese. Results show superior English performance by models like GPT series and Qwen series, while Chinese-centric LLMs excel in Chinese tests, albeit specialized IP LLMs lag behind general-purpose ones. Regional and temporal aspects of IP underscore the need for LLMs to grasp legal nuances and evolving laws. IPEval aims to accurately gauge LLM capabilities in IP and spur development of specialized models. Website: https://ipeval.github.io/

While text-to-image generation has been extensively studied, generating images from scene graphs remains relatively underexplored, primarily due to challenges in accurately modeling spatial relationships and object interactions. To fill this gap, we introduce Scene-Bench, a comprehensive benchmark designed to evaluate and enhance the factual consistency in generating natural scenes. Scene-Bench comprises MegaSG, a large-scale dataset of one million images annotated with scene graphs, facilitating the training and fair comparison of models across diverse and complex scenes. Additionally, we propose SGScore, a novel evaluation metric that leverages chain-of-thought reasoning capabilities of multimodal large language models (LLMs) to assess both object presence and relationship accuracy, offering a more effective measure of factual consistency than traditional metrics like FID and CLIPScore. Building upon this evaluation framework, we develop a scene graph feedback pipeline that iteratively refines generated images by identifying and correcting discrepancies between the scene graph and the image. Extensive experiments demonstrate that Scene-Bench provides a more comprehensive and effective evaluation framework compared to existing benchmarks, particularly for complex scene generation. Furthermore, our feedback strategy significantly enhances the factual consistency of image generation models, advancing the field of controllable image generation.

Large language models (LLMs) have demonstrated exceptional performance in various natural language processing tasks, yet their efficacy in more challenging and domain-specific tasks remains largely unexplored. This paper presents FinEval, a benchmark specifically designed for the financial domain knowledge in the LLMs. FinEval is a collection of high-quality multiple-choice questions covering Finance, Economy, Accounting, and Certificate. It includes 4,661 questions spanning 34 different academic subjects. To ensure a comprehensive model performance evaluation, FinEval employs a range of prompt types, including zero-shot and few-shot prompts, as well as answer-only and chain-of-thought prompts. Evaluating state-of-the-art Chinese and English LLMs on FinEval, the results show that only GPT-4 achieved an accuracy close to 70% in different prompt settings, indicating significant growth potential for LLMs in the financial domain knowledge. Our work offers a more comprehensive financial knowledge evaluation benchmark, utilizing data of mock exams and covering a wide range of evaluated LLMs.

Precisely evaluating video understanding models remains challenging: commonly used metrics such as BLEU, ROUGE, and BERTScore fail to capture the fineness of human judgment, while obtaining such judgments through manual evaluation is costly. Recent work has explored using large language models (LLMs) or multimodal LLMs (MLLMs) as evaluators, but their extension to video understanding remains relatively unexplored. In this work, we introduce VideoJudge, a 3B and 7B-sized MLLM judge specialized to evaluate outputs from video understanding models (i.e., text responses conditioned on videos). To train VideoJudge, our recipe builds on the interplay between a generator and an evaluator: the generator is prompted to produce responses conditioned on a target rating, and responses not matching the evaluator's rating are discarded. Across three out of four meta-evaluation benchmarks, VideoJudge-7B outperforms larger MLLM judge baselines such as Qwen2.5-VL (32B and 72B). Notably, we find that LLM judges (Qwen3) models perform worse than MLLM judges (Qwen2.5-VL) and long chain-of-thought reasoning does not improve performance, indicating that providing video inputs is crucial for evaluation of video understanding tasks.

Multi-agent debate (MAD) has gained significant attention as a promising line of research to improve the factual accuracy and reasoning capabilities of large language models (LLMs). Despite its conceptual appeal, current MAD research suffers from critical limitations in evaluation practices, including limited benchmark coverage, weak baseline comparisons, and inconsistent setups. This paper presents a systematic evaluation of 5 representative MAD methods across 9 benchmarks using 4 foundational models. Surprisingly, our findings reveal that MAD often fail to outperform simple single-agent baselines such as Chain-of-Thought and Self-Consistency, even when consuming significantly more inference-time computation. To advance MAD research, we further explore the role of model heterogeneity and find it as a universal antidote to consistently improve current MAD frameworks. Based on our findings, we argue that the field must stop overvaluing MAD in its current form; for true advancement, we must critically rethink evaluation paradigms and actively embrace model heterogeneity as a core design principle.

Driven by the remarkable progress in diffusion models, text-to-image generation has made significant strides, creating a pressing demand for automatic quality evaluation of generated images. Current state-of-the-art automatic evaluation methods heavily rely on Multi-modal Large Language Models (MLLMs), particularly powerful commercial models like GPT-4o. While these models are highly effective, their substantial costs limit scalability in large-scale evaluations. Adopting open-source MLLMs is an alternative; however, their performance falls short due to significant limitations in processing multi-modal data compared to commercial MLLMs. To tackle these problems, we first propose a task decomposition evaluation framework based on GPT-4o to automatically construct a new training dataset, where the complex evaluation task is decoupled into simpler sub-tasks, effectively reducing the learning complexity. Based on this dataset, we design innovative training strategies to effectively distill GPT-4o's evaluation capabilities into a 7B open-source MLLM, MiniCPM-V-2.6. Furthermore, to reliably and comprehensively assess prior works and our proposed model, we manually annotate a meta-evaluation benchmark that includes chain-of-thought explanations alongside quality scores for generated images. Experimental results demonstrate that our distilled open-source MLLM significantly outperforms the current state-of-the-art GPT-4o-base baseline, VIEScore, with over 4.6\% improvement in Spearman and Kendall correlations with human judgments.

Alignment has become a critical step for instruction-tuned Large Language Models (LLMs) to become helpful assistants. However, effective evaluation of alignment for emerging Chinese LLMs is still significantly lacking, calling for real-scenario grounded, open-ended, challenging and automatic evaluations tailored for alignment. To fill in this gap, we introduce AlignBench, a comprehensive multi-dimensional benchmark for evaluating LLMs' alignment in Chinese. Equipped with a human-in-the-loop data curation pipeline, our benchmark employs a rule-calibrated multi-dimensional LLM-as-Judge with Chain-of-Thought to generate explanations and final ratings as evaluations, ensuring high reliability and interpretability. Furthermore, we report AlignBench evaluated by CritiqueLLM, a dedicated Chinese evaluator LLM that recovers 95% of GPT-4's evaluation ability. We will provide public APIs for evaluating AlignBench with CritiqueLLM to facilitate the evaluation of LLMs' Chinese alignment. All evaluation codes, data, and LLM generations are available at https://github.com/THUDM/AlignBench.