多模态85 个 · 均热 18.1MMMU热 54.4Video-MME热 51.4fleurs热 48.6olmOCR-bench热 41MathVista热 40DocVQA热 39.9MMMU-Pro热 36.7speech commands热 36.6BLINK热 35.3MusicCaps热 35.2GenAI-Bench热 33.5CharXiv热 32Food Portion Benchmark热 30.6M-BEIR热 30MMEB-eval热 28.8Video-MME-v2热 28.6
垂直领域57 个 · 均热 14.7MedQA热 38.2+0.1LegalBench热 36-0.4frontierscience热 27.9multi lexsum热 27.1LEXam热 27FinanceBench热 26.7scicite热 25.7CritPt热 25.2peer read热 25HealthBench热 22.4climate-evaluation热 22.2novalad-evaluation热 22.2
代码51 个 · 均热 20.7HumanEval热 61.5-0.4SWE-bench Verified热 58.9-1.2MBPP热 56.1+1SWE-bench Lite热 49.9-0.1code contests热 45.6SWE-bench热 44.5LiveCodeBench热 43BigCodeBench热 40.3code x glue ct code to text热 36.5humanevalpack热 33.8code x glue cc code to code trans热 31code x glue cc code refinement热 30.2
智能体48 个 · 均热 10.6GAIA热 46.2-0.6GDPval热 37.6-0.2ClawBench热 31.1+0.4RoboDojo热 30.5+0.6GAIA-2热 29.3-0.2deepsearchqa热 28.4-0.6NexusRaven API evaluation热 27.3+0.1OSWorld热 27τ²-bench热 24.3
推理46 个 · 均热 15.1bigbench热 34.4cosmos qa热 34-0.2swag热 33.9+0.1ropes热 33.4-0.1art热 31.8+0.2winogrande热 29.4-0.4quoref热 28.5-0.4wiqa热 25.5social i qa热 24.2
安全对齐39 个 · 均热 16.7real-toxicity-prompts热 41.5+0.2wildjailbreak热 36.4-0.7WMDP热 36.2+0.1civil comments热 35.5-0.5AgentHarm热 32.9+0.2openai-moderation-api-evaluation热 32.1-0.3coconot热 30.7-0.3BeaverTails-Evaluation热 29.7-0.4prosocial-dialog热 28.8-0.3
语言/语义29 个 · 均热 15.2xtreme热 38.8+0.3lm1b热 32.1-0.1AlGhafa-Arabic-LLM-Benchmark-Native热 31+0.1snips built in intents热 30.2-0.2IndicGenBench flores in热 23.9-0.2Instruction-Following-Evaluation-for-Large-Language-Models热 23.9+0.2
知识问答28 个 · 均热 30.6MMLU-Pro热 66-0.1ai2 arc热 54.7+0.4MMLU热 51.5-0.2GPQA Diamond热 46-0.1TruthfulQA热 45.6+0.1sciq热 45.2-2.9