AI model builders mostly highlight unique benchmarks that act as flexible narrative tools for market positioning rather than standardized scientific measurements.
hub Mixed citations
MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI
Mixed citation behavior. Most common role is background (53%).
60
Pith papers citing it
Background
53% of classified citations
abstract
We introduce MMMU: a new benchmark designed to evaluate multimodal models on massive multi-discipline tasks demanding college-level subject knowledge and deliberate reasoning. MMMU includes 11.5K meticulously collected multimodal questions from college exams, quizzes, and textbooks, covering six core disciplines: Art & Design, Business, Science, Health & Medicine, Humanities & Social Science, and Tech & Engineering. These questions span 30 subjects and 183 subfields, comprising 30 highly heterogeneous image types, such as charts, diagrams, maps, tables, music sheets, and chemical structures. Unlike existing benchmarks, MMMU focuses on advanced perception and reasoning with domain-specific knowledge, challenging models to perform tasks akin to those faced by experts. The evaluation of 14 open-source LMMs as well as the proprietary GPT-4V(ision) and Gemini highlights the substantial challenges posed by MMMU. Even the advanced GPT-4V and Gemini Ultra only achieve accuracies of 56% and 59% respectively, indicating significant room for improvement. We believe MMMU will stimulate the community to build next-generation multimodal foundation models towards expert artificial general intelligence.
hub tools
citation-role summary
background 9
dataset 5
baseline 1
citation-polarity summary
representative citing papers
The SDE benchmark shows LLMs lag on scientific discovery tasks relative to general science tests, with diminishing scaling returns and shared weaknesses across models.
Introduces MMBU benchmark for VLMs in biomedicine and demonstrates that established benchmarks mask perception deficiencies in evaluated models.
UltraVR is a new diagnostic benchmark for evidence-grounded VQA on ultra-resolution images, with structured chain-of-thought annotations that localize failures in grounding, perception, and inference.
Visual debiasing of omni-modal benchmarks combined with staged post-training lets a 3B model match or exceed a 30B model without a stronger teacher.
PRISM-VL improves VLM performance by grounding on RAW-derived Meas.-XYZ inputs and exposure-bracketed supervision, gaining +0.1074 BLEU and +4.46% LLM-Judge accuracy over an RGB baseline on a held-out benchmark.
Medical VLMs frequently select negated options that contradict visible chest X-ray findings, achieving only ~30% accuracy on direct presence probes, but a post-hoc consistency verifier raises accuracy above 95%.
ReaLB balances multimodal MoE inference loads by switching vision-heavy experts to lower FP4 precision per device rank, hiding the change in the dispatch phase to deliver 1.10-1.32x speedup with <1% accuracy degradation.
PlotChain benchmark reports top MLLMs reaching ~80% field-level accuracy on engineering plot reading under human-like tolerances, but with persistent failures on frequency-domain tasks like bandpass and FFT spectra.
FinCriticalED benchmark reveals that OCR and MLLM systems frequently fail to preserve critical financial facts such as numbers and monetary units even when lexical accuracy is high.
FLARE is a vision-language model family using text-guided vision encoding, context-aware alignment decoding, dual-semantic mapping loss, and text-driven VQA synthesis to achieve deep cross-modal integration, outperforming larger models with only 630 vision tokens at 3B scale.
PolyMATH is a new 5,000-image benchmark where top MLLMs reach at most 41 percent accuracy on multi-modal mathematical reasoning, with ablation showing minimal gain from text over images.
V-RoAst applies zero-shot VLMs (Gemini-1.5-flash, GPT-4o-mini) to iRAP road safety attribute classification on a new ThaiRAP image dataset and compares them to CNN baselines, finding better generalization to unseen classes but weaker spatial reasoning.
LLaVA-NeXT-Interleave unifies multi-image, video, and 3D capabilities in large multimodal models via a new 1.18M-sample interleaved dataset and benchmark, achieving leading results across those tasks while preserving single-image performance.
WE-MATH benchmark reveals most LMMs rely on rote memorization for visual math while GPT-4o has shifted toward knowledge generalization.
MuirBench is a new benchmark showing that top multimodal LLMs struggle with robust multi-image understanding, with GPT-4o at 68% and open-source models below 33% accuracy.
MLVU is a new benchmark for long video understanding that uses extended videos across diverse genres and multi-task evaluations, revealing that current MLLMs struggle significantly and degrade sharply with longer durations.
MathVerse is a benchmark that tests multi-modal LLMs on visual math by providing each problem in six versions with progressively less diagram and text information to measure true visual understanding.
HPP decouples perception from reasoning in long-video VLMs by having an LLM run iterative programmatic probes on hierarchically segmented video, reporting gains on LongVideoBench, EgoSchema, VideoMME, and MLVU.
QK Product Steering suppresses dominant singular modes in the per-head QK product of selected middle layers via a closed-form query-only update, yielding 4.0% average relative CHAIR_s reduction on three GQA VLMs.
Qwen-RobotWorld is a language-conditioned video world model using Double-Stream MMDiT, an 8.6M-frame embodied corpus, and progressive curriculum training that ranks first on EWMBench and DreamGen Bench.
MOSS-Video-Preview introduces a cross-attention architecture and synthesized real-time QA data to enable continuous perception, answer revision, and faster inference in video-language models compared to decoder-only designs.
AsyMoE adds hyperbolic geometry for cross-modal hierarchies and evidence-priority experts to address vision-language asymmetry in LVLMs, reporting 1.5% average gains and 25.45% fewer active parameters.
MMTB is a new benchmark with 105 multimedia terminal tasks that shows how audio and video access changes agent performance and evidence use in executable workflows.
citing papers explorer
-
Unsteady Metrics and Benchmarking Cultures of AI Model Builders
AI model builders mostly highlight unique benchmarks that act as flexible narrative tools for market positioning rather than standardized scientific measurements.
-
Evaluating Large Language Models in Scientific Discovery
The SDE benchmark shows LLMs lag on scientific discovery tasks relative to general science tests, with diminishing scaling returns and shared weaknesses across models.
-
MMBU: A Massive Multi-modal Biomedical Understanding Benchmark to Probe the Perception Capabilities of Vision-Language Models
Introduces MMBU benchmark for VLMs in biomedicine and demonstrates that established benchmarks mask perception deficiencies in evaluated models.
-
UltraVR: A Diagnostic Ultra-Resolution Image-VQA Benchmark for Evidence-Grounded Reasoning
UltraVR is a new diagnostic benchmark for evidence-grounded VQA on ultra-resolution images, with structured chain-of-thought annotations that localize failures in grounding, perception, and inference.
-
Boosting Omni-Modal Language Models: Staged Post-Training with Visually Debiased Evaluation
Visual debiasing of omni-modal benchmarks combined with staged post-training lets a 3B model match or exceed a 30B model without a stronger teacher.
-
Allegory of the Cave: Measurement-Grounded Vision-Language Learning
PRISM-VL improves VLM performance by grounding on RAW-derived Meas.-XYZ inputs and exposure-bracketed supervision, gaining +0.1074 BLEU and +4.46% LLM-Judge accuracy over an RGB baseline on a held-out benchmark.
-
CXR-ContraBench: Benchmarking Negated-Option Attraction in Medical VLMs
Medical VLMs frequently select negated options that contradict visible chest X-ray findings, achieving only ~30% accuracy on direct presence probes, but a post-hoc consistency verifier raises accuracy above 95%.
-
ReaLB: Real-Time Load Balancing for Multimodal MoE Inference
ReaLB balances multimodal MoE inference loads by switching vision-heavy experts to lower FP4 precision per device rank, hiding the change in the dispatch phase to deliver 1.10-1.32x speedup with <1% accuracy degradation.
-
PlotChain: Deterministic Checkpointed Evaluation of Multimodal LLMs on Engineering Plot Reading
PlotChain benchmark reports top MLLMs reaching ~80% field-level accuracy on engineering plot reading under human-like tolerances, but with persistent failures on frequency-domain tasks like bandpass and FFT spectra.
-
FinCriticalED: A Visual Benchmark for Financial Fact-Level OCR
FinCriticalED benchmark reveals that OCR and MLLM systems frequently fail to preserve critical financial facts such as numbers and monetary units even when lexical accuracy is high.
-
FLARE: Fully Integration of Vision-Language Representations for Deep Cross-Modal Understanding
FLARE is a vision-language model family using text-guided vision encoding, context-aware alignment decoding, dual-semantic mapping loss, and text-driven VQA synthesis to achieve deep cross-modal integration, outperforming larger models with only 630 vision tokens at 3B scale.
-
Polymath: A Challenging Multi-modal Mathematical Reasoning Benchmark
PolyMATH is a new 5,000-image benchmark where top MLLMs reach at most 41 percent accuracy on multi-modal mathematical reasoning, with ablation showing minimal gain from text over images.
-
V-RoAst: Visual Road Assessment. Can VLM be a Road Safety Assessor Using the iRAP Standard?
V-RoAst applies zero-shot VLMs (Gemini-1.5-flash, GPT-4o-mini) to iRAP road safety attribute classification on a new ThaiRAP image dataset and compares them to CNN baselines, finding better generalization to unseen classes but weaker spatial reasoning.
-
LLaVA-NeXT-Interleave: Tackling Multi-image, Video, and 3D in Large Multimodal Models
LLaVA-NeXT-Interleave unifies multi-image, video, and 3D capabilities in large multimodal models via a new 1.18M-sample interleaved dataset and benchmark, achieving leading results across those tasks while preserving single-image performance.
-
We-Math: Does Your Large Multimodal Model Achieve Human-like Mathematical Reasoning?
WE-MATH benchmark reveals most LMMs rely on rote memorization for visual math while GPT-4o has shifted toward knowledge generalization.
-
MuirBench: A Comprehensive Benchmark for Robust Multi-image Understanding
MuirBench is a new benchmark showing that top multimodal LLMs struggle with robust multi-image understanding, with GPT-4o at 68% and open-source models below 33% accuracy.
-
MLVU: Benchmarking Multi-task Long Video Understanding
MLVU is a new benchmark for long video understanding that uses extended videos across diverse genres and multi-task evaluations, revealing that current MLLMs struggle significantly and degrade sharply with longer durations.
-
MathVerse: Does Your Multi-modal LLM Truly See the Diagrams in Visual Math Problems?
MathVerse is a benchmark that tests multi-modal LLMs on visual math by providing each problem in six versions with progressively less diagram and text information to measure true visual understanding.
-
HPP: Hierarchical Programmatic Probing for Long Video Understanding by Decoupling Perception and Reasoning
HPP decouples perception from reasoning in long-video VLMs by having an LLM run iterative programmatic probes on hierarchically segmented video, reporting gains on LongVideoBench, EgoSchema, VideoMME, and MLVU.
-
Spectral Query-Key Product Weight Steering for Training-Free VLM Hallucination Mitigation
QK Product Steering suppresses dominant singular modes in the per-head QK product of selected middle layers via a closed-form query-only update, yielding 4.0% average relative CHAIR_s reduction on three GQA VLMs.
-
Qwen-RobotWorld Technical Report: Unifying Embodied World Modeling through Language-Conditioned Video Generation
Qwen-RobotWorld is a language-conditioned video world model using Double-Stream MMDiT, an 8.6M-frame embodied corpus, and progressive curriculum training that ranks first on EWMBench and DreamGen Bench.
-
MOSS-Video-Preview: Toward Real-Time Video Understanding via Cross-Attention
MOSS-Video-Preview introduces a cross-attention architecture and synthesized real-time QA data to enable continuous perception, answer revision, and faster inference in video-language models compared to decoder-only designs.
-
Hyperbolic and Evidence-Prioritized Experts for Large Vision-Language Models
AsyMoE adds hyperbolic geometry for cross-modal hierarchies and evidence-priority experts to address vision-language asymmetry in LVLMs, reporting 1.5% average gains and 25.45% fewer active parameters.
-
MMTB: Evaluating Terminal Agents on Multimedia-File Tasks
MMTB is a new benchmark with 105 multimedia terminal tasks that shows how audio and video access changes agent performance and evidence use in executable workflows.
-
Co-Evolving Policy Distillation
CoPD integrates multiple expert capabilities by running parallel RLVR training with bidirectional online policy distillation among experts, outperforming mixed RLVR and sequential OPD while surpassing domain-specific experts on text-image-video reasoning.
-
Agentic Learner with Grow-and-Refine Multimodal Semantic Memory
ViLoMem is a dual-stream grow-and-refine memory system that separates visual and logical error patterns in MLLMs to improve pass@1 accuracy and reduce repeated mistakes across six multimodal benchmarks.
-
Qwen3-Omni Technical Report
Qwen3-Omni is a unified multimodal model that achieves open-source SOTA on 32 of 36 audio and audio-visual benchmarks and overall SOTA on 22 without degrading performance on text, image, or video relative to single-modal Qwen counterparts.
-
InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency
InternVL3.5 advances open-source multimodal models with Cascade RL for +16% reasoning gains and ViR for 4x inference speedup, with the 241B model reaching SOTA among open-source MLLMs on multimodal, reasoning, and agentic tasks.
-
Fourier Compressor: Frequency-Domain Visual Token Compression for Vision-Language Models
Fourier Compressor uses FFT to remove frequency-domain redundancy from visual tokens in VLMs, retaining over 96% accuracy with up to 83.8% FLOP reduction.
-
Beyond the Leaderboard: Rethinking Medical Benchmarks for Large Language Models
MedCheck is a lifecycle checklist framework that audits 53 existing medical LLM benchmarks and identifies systemic gaps in clinical fidelity, contamination control, and safety metrics.
-
Grounded Reinforcement Learning for Visual Reasoning
ViGoRL introduces visually grounded RL that anchors reasoning steps to image coordinates and uses multi-turn zooming to outperform standard RL and supervised baselines on spatial and GUI reasoning benchmarks.
-
InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models
InternVL3-78B sets a new open-source SOTA of 72.2 on MMMU via native joint multimodal pre-training, V2PE, MPO, and test-time scaling while remaining competitive with proprietary models.
-
Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling
InternVL 2.5 is the first open-source MLLM to surpass 70% on the MMMU benchmark via model, data, and test-time scaling, with a 3.7-point gain from chain-of-thought reasoning.
-
Capabilities of Gemini Models in Medicine
Med-Gemini sets new records on 10 of 14 medical benchmarks including 91.1% on MedQA-USMLE, beats GPT-4V by 44.5% on multimodal tasks, and surpasses humans on medical text summarization.
-
BLINK: Multimodal Large Language Models Can See but Not Perceive
BLINK benchmark shows multimodal LLMs reach only 45-51 percent accuracy on core visual perception tasks where humans achieve 95 percent, indicating these abilities have not emerged.
-
Are We on the Right Way for Evaluating Large Vision-Language Models?
Current LVLM benchmarks overestimate capabilities because many questions can be answered without images due to design flaws or data leakage; MMStar is a human-curated set of 1,500 vision-indispensable samples across 6 capabilities and 18 axes with new metrics for leakage and true multi-modal gain.
-
MM1: Methods, Analysis & Insights from Multimodal LLM Pre-training
MM1 models achieve state-of-the-art few-shot multimodal results by pre-training on a careful mix of image-caption, interleaved, and text-only data with optimized image encoders.
-
Measuring Multimodal Mathematical Reasoning with MATH-Vision Dataset
MATH-Vision is a new benchmark of 3,040 visual mathematical competition problems that reveals substantial gaps between large multimodal models and human performance in mathematical reasoning.
-
ALLaVA: Harnessing GPT4V-Synthesized Data for Lite Vision-Language Models
ALLaVA creates 1.3M GPT4V-synthesized samples enabling 4B VLMs to achieve competitive results on 17 benchmarks and match 7B/13B models on some tasks.
-
GPT-4V(ision) is a Generalist Web Agent, if Grounded
GPT-4V achieves 51.1% success on live web tasks as a generalist agent when plans are manually grounded, outperforming text-only models, but automatic grounding lags far behind oracle performance.
-
CogVLM: Visual Expert for Pretrained Language Models
CogVLM adds a trainable visual expert inside frozen language model layers for deep vision-language fusion and reports state-of-the-art results on ten cross-modal benchmarks while preserving NLP performance.
-
ESC: Emotional Self-Correction for Reliable Vision-Language Models
ESC uses emotional cues triggered by an external verifier to enable training-free self-correction in VLMs, improving reliability on safety, hallucination, and reasoning benchmarks.
-
TuringViT: Making SOTA Vision Transformers Accessible to All
TuringViT claims a new ViT design with linear attention and curated data that matches SOTA performance using 10% of typical pretraining data while supporting dynamic resolutions and improving VLM integration.
-
SingGuard: A Policy-Adaptive Multimodal LLM Guardrail with Dynamic Reasoning
SingGuard introduces a policy-adaptive multimodal LLM guardrail with dynamic reasoning regimes and SingGuard-Bench, reporting SOTA F1 scores across 35 datasets and improved policy-following accuracy under runtime shifts.
-
Modeling Complex Behaviors: Multi-Personality Composition and Dynamic Switching in Vision-Language Models
The work establishes an evaluation framework for personality induction and switching in MLLMs, reporting improved captioning but impaired VQA performance plus balancing and residual effects during multi-trait and dynamic conditions.
-
Do Vision-Language Models See Dwarf Galaxies the Way We Do?
Zero-shot VLMs reproduce aggregate human annotations on dwarf galaxy detection but exhibit high per-example variability and unreliable self-reported confidence.
-
PhyDrawGen: Physically Grounded Diagram Generation from Natural Language
PhyDrawGen is a neuro-symbolic pipeline that extracts typed scene graphs via LLM, converts them to physically constrained PSLGs via deterministic solver, and refines via fine-tuned Qwen-VL, claiming superior performance over GPT-5-image and Gemini models on 1,449 physics problems.
-
AlphaEval: Evaluating Agents in Production
AlphaEval is a benchmark of 94 production-sourced tasks from seven companies for evaluating full AI agent products across six domains using multiple judgment methods, plus a framework to build similar benchmarks.
-
Back to the Barn with LLAMAs: Evolving Pretrained LLM Backbones in Finetuning Vision Language Models
Newer LLM backbones in VLMs do not always improve performance; gains are task-dependent, with VQA models solving different questions due to better confidence calibration and stable representations.
-
AVATAAR: Agentic Video Answering via Temporal Adaptive Alignment and Reasoning
AVATAAR reports relative gains of 5-8% over baseline on CinePile benchmark categories through agentic feedback for long video QA.