Large Language Models (LLMs) have demonstrated exceptional capabilities across diverse natural language processing (NLP) tasks. The release of open-source LLMs like LLaMA and Qwen has triggered the development of numerous fine-tuned models tailored for various tasks and languages. In this paper, we explore an important question: is it possible to combine these specialized models to create a unified model with multi-task capabilities. We introduces Hierarchical Iterative Merging (Hi-Merging), a training-free method for unifying different specialized LLMs into a single model. Specifically, Hi-Merging employs model-wise and layer-wise pruning and scaling, guided by contribution analysis, to mitigate parameter conflicts. Extensive experiments on multiple-choice and question-answering tasks in both Chinese and English validate Hi-Merging's ability for multi-task learning. The results demonstrate that Hi-Merging consistently outperforms existing merging techniques and surpasses the performance of models fine-tuned on combined datasets in most scenarios. Code is available at: https://github.com/Applied-Machine-Learning-Lab/Hi-Merging.
We introduce STSBench, a scenario-based framework to benchmark the holistic understanding of vision-language models (VLMs) for autonomous driving. The framework automatically mines pre-defined traffic scenarios from any dataset using ground-truth annotations, provides an intuitive user interface for efficient human verification, and generates multiple-choice questions for model evaluation. Applied to the NuScenes dataset, we present STSnu, the first benchmark that evaluates the spatio-temporal reasoning capabilities of VLMs based on comprehensive 3D perception. Existing benchmarks typically target off-the-shelf or fine-tuned VLMs for images or videos from a single viewpoint and focus on semantic tasks such as object recognition, dense captioning, risk assessment, or scene understanding. In contrast, STSnu evaluates driving expert VLMs for end-to-end driving, operating on videos from multi-view cameras or LiDAR. It specifically assesses their ability to reason about both ego-vehicle actions and complex interactions among traffic participants, a crucial capability for autonomous vehicles. The benchmark features 43 diverse scenarios spanning multiple views and frames, resulting in 971 human-verified multiple-choice questions. A thorough evaluation uncovers critical shortcomings in existing models' ability to reason about fundamental traffic dynamics in complex environments. These findings highlight the urgent need for architectural advances that explicitly model spatio-temporal reasoning. By addressing a core gap in spatio-temporal evaluation, STSBench enables the development of more robust and explainable VLMs for autonomous driving.
Evaluating large language models (LLMs) in medicine is crucial because medical applications require high accuracy with little room for error. Current medical benchmarks have three main types: medical exam-based, comprehensive medical, and specialized assessments. However, these benchmarks have limitations in question design (mostly multiple-choice), data sources (often not derived from real clinical scenarios), and evaluation methods (poor assessment of complex reasoning). To address these issues, we present LLMEval-Med, a new benchmark covering five core medical areas, including 2,996 questions created from real-world electronic health records and expert-designed clinical scenarios. We also design an automated evaluation pipeline, incorporating expert-developed checklists into our LLM-as-Judge framework. Furthermore, our methodology validates machine scoring through human-machine agreement analysis, dynamically refining checklists and prompts based on expert feedback to ensure reliability. We evaluate 13 LLMs across three categories (specialized medical models, open-source models, and closed-source models) on LLMEval-Med, providing valuable insights for the safe and effective deployment of LLMs in medical domains. The dataset is released in https://github.com/llmeval/LLMEval-Med.
We introduce the $\underline{Ko}rean \underline{G}rammar \underline{E}valuation Bench\underline{M}ark (KoGEM)$, designed to assess the linguistic competence of LLMs and humans in Korean. KoGEM consists of 1.5k multiple-choice QA pairs covering five main categories and 16 subcategories. The zero-shot evaluation of 27 LLMs of various sizes and types reveals that while LLMs perform remarkably well on straightforward tasks requiring primarily definitional knowledge, they struggle with tasks that demand the integration of real-world experiential knowledge, such as phonological rules and pronunciation. Furthermore, our in-depth analysis suggests that incorporating such experiential knowledge could enhance the linguistic competence of LLMs. With KoGEM, we not only highlight the limitations of current LLMs in linguistic competence but also uncover hidden facets of LLMs in linguistic competence, paving the way for enhancing comprehensive language understanding. Our code and dataset are available at: https://github.com/SungHo3268/KoGEM.
The performance of large language models (LLMs) continues to improve, as reflected in rising scores on standard benchmarks. However, the lack of transparency around training data raises concerns about potential overlap with evaluation sets and the fairness of reported results. Although prior work has proposed methods for detecting data leakage, these approaches primarily focus on identifying outliers and have not been evaluated under controlled simulated leakage conditions. In this work, we compare existing leakage detection techniques, namely permutation and n-gram-based methods, under a continual pretraining setup that simulates real-world leakage scenarios, and additionally explore a lightweight method we call semi-half question. Although semi-half offers a low-cost alternative, our analysis shows that the n-gram method consistently achieves the highest F1-score. We also refine these techniques to support instance-level detection and reduce computational overhead. Leveraging the best-performing method, we create cleaned versions of MMLU and HellaSwag, and re-evaluate several LLMs. Our findings present a practical path toward more reliable and transparent evaluations, and we recommend contamination checks as a standard step before releasing benchmark results.
Recent works on large language models (LLMs) have successfully demonstrated the emergence of reasoning capabilities via reinforcement learning (RL). Although recent efforts leverage group relative policy optimization (GRPO) for MLLMs post-training, they constantly explore one specific aspect, such as grounding tasks, math problems, or chart analysis. There are no works that can leverage multi-source MLLM tasks for stable reinforcement learning. In this work, we present a unified perspective to solve this problem. We present Mixed-R1, a unified yet straightforward framework that contains a mixed reward function design (Mixed-Reward) and a mixed post-training dataset (Mixed-45K). We first design a data engine to select high-quality examples to build the Mixed-45K post-training dataset. Then, we present a Mixed-Reward design, which contains various reward functions for various MLLM tasks. In particular, it has four different reward functions: matching reward for binary answer or multiple-choice problems, chart reward for chart-aware datasets, IoU reward for grounding problems, and open-ended reward for long-form text responses such as caption datasets. To handle the various long-form text content, we propose a new open-ended reward named Bidirectional Max-Average Similarity (BMAS) by leveraging tokenizer embedding matching between the generated response and the ground truth. Extensive experiments show the effectiveness of our proposed method on various MLLMs, including Qwen2.5-VL and Intern-VL on various sizes. Our dataset and model are available at https://github.com/xushilin1/mixed-r1.
With the increasing integration of visual and textual content in Social Networking Services (SNS), evaluating the multimodal capabilities of Large Language Models (LLMs) is crucial for enhancing user experience, content understanding, and platform intelligence. Existing benchmarks primarily focus on text-centric tasks, lacking coverage of the multimodal contexts prevalent in modern SNS ecosystems. In this paper, we introduce SNS-Bench-VL, a comprehensive multimodal benchmark designed to assess the performance of Vision-Language LLMs in real-world social media scenarios. SNS-Bench-VL incorporates images and text across 8 multimodal tasks, including note comprehension, user engagement analysis, information retrieval, and personalized recommendation. It comprises 4,001 carefully curated multimodal question-answer pairs, covering single-choice, multiple-choice, and open-ended tasks. We evaluate over 25 state-of-the-art multimodal LLMs, analyzing their performance across tasks. Our findings highlight persistent challenges in multimodal social context comprehension. We hope SNS-Bench-VL will inspire future research towards robust, context-aware, and human-aligned multimodal intelligence for next-generation social networking services.
Open benchmarks are essential for evaluating and advancing large language models, offering reproducibility and transparency. However, their accessibility makes them likely targets of test set contamination. In this work, we introduce DyePack, a framework that leverages backdoor attacks to identify models that used benchmark test sets during training, without requiring access to the loss, logits, or any internal details of the model. Like how banks mix dye packs with their money to mark robbers, DyePack mixes backdoor samples with the test data to flag models that trained on it. We propose a principled design incorporating multiple backdoors with stochastic targets, enabling exact false positive rate (FPR) computation when flagging every model. This provably prevents false accusations while providing strong evidence for every detected case of contamination. We evaluate DyePack on five models across three datasets, covering both multiple-choice and open-ended generation tasks. For multiple-choice questions, it successfully detects all contaminated models with guaranteed FPRs as low as 0.000073% on MMLU-Pro and 0.000017% on Big-Bench-Hard using eight backdoors. For open-ended generation tasks, it generalizes well and identifies all contaminated models on Alpaca with a guaranteed false positive rate of just 0.127% using six backdoors.
Video Anomaly Understanding (VAU) is essential for applications such as smart cities, security surveillance, and disaster alert systems, yet remains challenging due to its demand for fine-grained spatio-temporal perception and robust reasoning under ambiguity. Despite advances in anomaly detection, existing methods often lack interpretability and struggle to capture the causal and contextual aspects of abnormal events. This limitation is further compounded by the absence of comprehensive benchmarks for evaluating reasoning ability in anomaly scenarios. To address both challenges, we introduce VAU-R1, a data-efficient framework built upon Multimodal Large Language Models (MLLMs), which enhances anomaly reasoning through Reinforcement Fine-Tuning (RFT). Besides, we propose VAU-Bench, the first Chain-of-Thought benchmark tailored for video anomaly reasoning, featuring multiple-choice QA, detailed rationales, temporal annotations, and descriptive captions. Empirical results show that VAU-R1 significantly improves question answering accuracy, temporal grounding, and reasoning coherence across diverse contexts. Together, our method and benchmark establish a strong foundation for interpretable and reasoning-aware video anomaly understanding. Our code is available at https://github.com/GVCLab/VAU-R1.
The Massive Multitask Language Understanding (MMLU) benchmark has been widely used to evaluate language models across various domains. However, existing MMLU datasets primarily focus on high-resource languages such as English, which leaves low-resource languages like Bengali underrepresented. In this paper, we introduce BnMMLU, a benchmark to evaluate the multitask language understanding capabilities of Bengali in language models. The dataset spans 23 domains, including science, humanities, mathematics and general knowledge and is structured in a multiple-choice format to assess factual knowledge, application-based problem-solving and reasoning abilities of language models. It consists of 138,949 question-option pairs. We benchmark several proprietary and open-source large language models (LLMs) on the BnMMLU test set. Additionally, we annotate the test set with three cognitive categories-factual knowledge, procedural application and reasoning-to gain deeper insights into model strengths and weaknesses across various cognitive tasks. The results reveal significant performance gaps, highlighting the need for improved pre-training and fine-tuning strategies tailored to Bengali data. We release the dataset and benchmark results to facilitate further research in this area.
Recent Multimodal Large Language Models (MLLMs) achieve promising performance on visual and audio benchmarks independently. However, the ability of these models to process cross-modal information synchronously remains largely unexplored. In this paper, we introduce: 1) Daily-Omni, an Audio-Visual Questioning and Answering benchmark comprising 684 videos of daily life scenarios from diverse sources, rich in both audio and visual information, and featuring 1197 multiple-choice QA pairs across 6 major tasks; 2) Daily-Omni QA Generation Pipeline, which includes automatic annotation, QA generation and QA optimization, significantly improves efficiency for human evaluation and scalability of the benchmark; 3) Daily-Omni-Agent, a training-free agent utilizing open-source Visual Language Model (VLM), Audio Language Model (ALM) and Automatic Speech Recognition (ASR) model to establish a baseline for this benchmark. The results show that current MLLMs still struggle significantly with tasks requiring audio-visual integration, but combining VLMs and ALMs with simple temporal alignment techniques can achieve substantially better performance. Codes and benchmark are available at \href{https://github.com/Lliar-liar/Daily-Omni}{https://github.com/Lliar-liar/Daily-Omni}.
Metaphorical comprehension in images remains a critical challenge for AI systems, as existing models struggle to grasp the nuanced cultural, emotional, and contextual implications embedded in visual content. While multimodal large language models (MLLMs) excel in basic Visual Question Answer (VQA) tasks, they struggle with a fundamental limitation on image implication tasks: contextual gaps that obscure the relationships between different visual elements and their abstract meanings. Inspired by the human cognitive process, we propose Let Androids Dream (LAD), a novel framework for image implication understanding and reasoning. LAD addresses contextual missing through the three-stage framework: (1) Perception: converting visual information into rich and multi-level textual representations, (2) Search: iteratively searching and integrating cross-domain knowledge to resolve ambiguity, and (3) Reasoning: generating context-alignment image implication via explicit reasoning. Our framework with the lightweight GPT-4o-mini model achieves SOTA performance compared to 15+ MLLMs on English image implication benchmark and a huge improvement on Chinese benchmark, performing comparable with the GPT-4o model on Multiple-Choice Question (MCQ) and outperforms 36.7% on Open-Style Question (OSQ). Additionally, our work provides new insights into how AI can more effectively interpret image implications, advancing the field of vision-language reasoning and human-AI interaction. Our project is publicly available at https://github.com/MING-ZCH/Let-Androids-Dream-of-Electric-Sheep.
Recent advances in Large Language Models (LLMs) have demonstrated promising knowledge and reasoning abilities, yet their performance in multilingual and low-resource settings remains underexplored. Existing benchmarks often exhibit cultural bias, restrict evaluation to text-only, rely on multiple-choice formats, and, more importantly, are limited for extremely low-resource languages. To address these gaps, we introduce IRLBench, presented in parallel English and Irish, which is considered definitely endangered by UNESCO. Our benchmark consists of 12 representative subjects developed from the 2024 Irish Leaving Certificate exams, enabling fine-grained analysis of model capabilities across domains. By framing the task as long-form generation and leveraging the official marking scheme, it does not only support a comprehensive evaluation of correctness but also language fidelity. Our extensive experiments of leading closed-source and open-source LLMs reveal a persistent performance gap between English and Irish, in which models produce valid Irish responses less than 80\% of the time, and answer correctly 55.8\% of the time compared to 76.2\% in English for the best-performing model. We release IRLBench (https://huggingface.co/datasets/ReliableAI/IRLBench) and an accompanying evaluation codebase (https://github.com/ReML-AI/IRLBench) to enable future research on robust, culturally aware multilingual AI development.
Recent progress in large language models (LLMs) has outpaced the development of effective evaluation methods. Traditional benchmarks rely on task-specific metrics and static datasets, which often suffer from fairness issues, limited scalability, and contamination risks. In this paper, we introduce Teach2Eval, an indirect evaluation framework inspired by the Feynman Technique. Instead of directly testing LLMs on predefined tasks, our method evaluates a model's multiple abilities to teach weaker student models to perform tasks effectively. By converting open-ended tasks into standardized multiple-choice questions (MCQs) through teacher-generated feedback, Teach2Eval enables scalable, automated, and multi-dimensional assessment. Our approach not only avoids data leakage and memorization but also captures a broad range of cognitive abilities that are orthogonal to current benchmarks. Experimental results across 26 leading LLMs show strong alignment with existing human and model-based dynamic rankings, while offering additional interpretability for training guidance.
Recent advances in Large Multimodal Models (LMMs) have significantly improved their reasoning and Optical Character Recognition (OCR) capabilities. However, their performance on complex logical reasoning tasks involving text-rich images remains underexplored. To bridge this gap, we introduce LogicOCR, a benchmark comprising 1,100 multiple-choice questions designed to evaluate LMMs' logical reasoning abilities on text-rich images, while minimizing reliance on domain-specific knowledge (e.g., mathematics). We construct LogicOCR by curating a text corpus from the Chinese National Civil Servant Examination and develop a scalable, automated pipeline to convert it into multimodal samples. First, we design prompt templates to steer GPT-Image-1 to generate images with diverse backgrounds, interleaved text-illustration layouts, and varied fonts, ensuring contextual relevance and visual realism. Then, the generated images are manually verified, with low-quality examples discarded. We evaluate a range of representative open-source and proprietary LMMs under both Chain-of-Thought (CoT) and direct-answer settings. Our multi-dimensional analysis reveals key insights, such as the impact of test-time scaling, input modality differences, and sensitivity to visual-text orientation. Notably, LMMs still lag in multimodal reasoning compared to text-only inputs, indicating that they have not fully bridged visual reading with reasoning. We hope LogicOCR will serve as a valuable resource for advancing multimodal reasoning research. The dataset is available at https://github.com/MiliLab/LogicOCR.
Majority voting is considered an effective method to enhance chain-of-thought reasoning, as it selects the answer with the highest "self-consistency" among different reasoning paths (Wang et al., 2023). However, previous chain-of-thought reasoning methods typically generate only a single answer in each trial, thereby ignoring the possibility of other potential answers. As a result, these alternative answers are often overlooked in subsequent voting processes. In this work, we propose to generate ranked answers in each reasoning process and conduct ranked voting among multiple ranked answers from different responses, thereby making the overall self-consistency more reliable. Specifically, we use three ranked voting methods: Instant-runoff voting, Borda count voting, and mean reciprocal rank voting. We validate our methods on six datasets, including three multiple-choice and three open-ended question-answering tasks, using both advanced open-source and closed-source large language models. Extensive experimental results indicate that our proposed method outperforms the baselines, showcasing the potential of leveraging the information of ranked answers and using ranked voting to improve reasoning performance. The code is available at https://github.com/szu-tera/RankedVotingSC.
Editing images using natural language instructions has become a natural and expressive way to modify visual content; yet, evaluating the performance of such models remains challenging. Existing evaluation approaches often rely on image-text similarity metrics like CLIP, which lack precision. In this work, we introduce a new benchmark designed to evaluate text-guided image editing models in a more grounded manner, along two critical dimensions: (i) functional correctness, assessed via automatically generated multiple-choice questions that verify whether the intended change was successfully applied; and (ii) image content preservation, which ensures that non-targeted regions of the image remain visually consistent using an object-aware masking technique and preservation scoring. The benchmark includes over 1000 high-quality editing examples across 20 diverse content categories, each annotated with detailed editing instructions, evaluation questions, and spatial object masks. We conduct a large-scale study comparing GPT-Image-1, the latest flagship in the text-guided image editing space, against several state-of-the-art editing models, and validate our automatic metrics against human ratings. Results show that GPT-Image-1 leads in instruction-following accuracy, but often over-modifies irrelevant image regions, highlighting a key trade-off in the current model behavior. GIE-Bench provides a scalable, reproducible framework for advancing more accurate evaluation of text-guided image editing.
Recent advances in vision language models (VLMs) have enabled broad progress in the general medical field. However, pathology still remains a more challenging subdomain, with current pathology specific VLMs exhibiting limitations in both diagnostic accuracy and reasoning plausibility. Such shortcomings are largely attributable to the nature of current pathology datasets, which are primarily composed of image description pairs that lack the depth and structured diagnostic paradigms employed by real world pathologists. In this study, we leverage pathology textbooks and real world pathology experts to construct high-quality, reasoning-oriented datasets. Building on this, we introduce Patho-R1, a multimodal RL-based pathology Reasoner, trained through a three-stage pipeline: (1) continued pretraining on 3.5 million image-text pairs for knowledge infusion; (2) supervised fine-tuning on 500k high-quality Chain-of-Thought samples for reasoning incentivizing; (3) reinforcement learning using Group Relative Policy Optimization and Decoupled Clip and Dynamic sAmpling Policy Optimization strategies for multimodal reasoning quality refinement. To further assess the alignment quality of our dataset, we propose PathoCLIP, trained on the same figure-caption corpus used for continued pretraining. Comprehensive experimental results demonstrate that both PathoCLIP and Patho-R1 achieve robust performance across a wide range of pathology-related tasks, including zero-shot classification, cross-modal retrieval, Visual Question Answering, and Multiple Choice Question. Our project is available at the Patho-R1 repository: https://github.com/Wenchuan-Zhang/Patho-R1.
Language Models (LMs) continue to advance, improving response quality and coherence. Given Internet-scale training datasets, LMs have likely encountered much of what users may ask them to generate in some form during their training. A plethora of evaluation benchmarks have been constructed to assess model quality, response appropriateness, and reasoning capabilities. However, the human effort required for benchmark construction is rapidly being outpaced by the size and scope of the models under evaluation. Having humans build a benchmark for every possible domain of interest is impractical. Therefore, we propose a methodology for automating the construction of fact-based synthetic data model evaluations grounded in document populations. This work leverages the same LMs to evaluate domain-specific knowledge automatically, using only grounding documents (e.g., a textbook) as input. This synthetic data benchmarking approach corresponds well with human curated questions producing a Spearman ranking correlation of 0.97 and a benchmark evaluation Pearson accuracy correlation of 0.75. This novel approach supports generating both multiple choice and open-ended synthetic data questions to gain diagnostic insight of LM capability. We apply this methodology to evaluate model performance on two recent arXiv preprints, discovering a surprisingly strong performance from Gemma-3 models on open-ended questions. Code is available at https://github.com/mmajurski/grounded-synth-lm-benchmark
We present HealthBench, an open-source benchmark measuring the performance and safety of large language models in healthcare. HealthBench consists of 5,000 multi-turn conversations between a model and an individual user or healthcare professional. Responses are evaluated using conversation-specific rubrics created by 262 physicians. Unlike previous multiple-choice or short-answer benchmarks, HealthBench enables realistic, open-ended evaluation through 48,562 unique rubric criteria spanning several health contexts (e.g., emergencies, transforming clinical data, global health) and behavioral dimensions (e.g., accuracy, instruction following, communication). HealthBench performance over the last two years reflects steady initial progress (compare GPT-3.5 Turbo's 16% to GPT-4o's 32%) and more rapid recent improvements (o3 scores 60%). Smaller models have especially improved: GPT-4.1 nano outperforms GPT-4o and is 25 times cheaper. We additionally release two HealthBench variations: HealthBench Consensus, which includes 34 particularly important dimensions of model behavior validated via physician consensus, and HealthBench Hard, where the current top score is 32%. We hope that HealthBench grounds progress towards model development and applications that benefit human health.