Skip to content

Benchmarks

Browse 5,586 benchmarks across 21 topics. Search by name or choose a topic to see the model rankings. Compare scores within the same benchmark.

Topics

Topics whose main benchmarks measure the most models come first. Open one for up to 10 of its most-measured main benchmarks.

Human preference136 benchmarksArena ELO, Arena ELO - English, Arena ELO - Exclude Ties…
Coding299 benchmarksArena ELO - Coding, Dtbench EDT answer preference, Arena ELO Text Factuality - Coding…
Instruction following38 benchmarksArena ELO - Instruction Following, IFEval, Arena ELO Text Factuality - Instruction Following…
Reasoning227 benchmarksSnakeBench, BBH, MuSR…
Math147 benchmarksArena ELO - Math, OTIS Mock AIME, Arena ELO Text Factuality - Math…
General105 benchmarksEpoch Capabilities Index, Dtbench Accuracy, Dtbench CRI-rescaled score…
Science60 benchmarksGPQA, CritPT, Vals GPQA Diamond…
Knowledge142 benchmarksMMLU-Pro, MMLU, Vals MMLU Pro…
Multilingual1,217 benchmarksEKI Estonian bib_bench, EKI Estonian idiom_bench, EKI Estonian keelenou…
Agentic379 benchmarkstau2-bench airline, SWE-Bench Pro, TBench…
Factuality59 benchmarksSimpleQA Verified, Vectara Factual Consistency, Vectara Hallucination Rate…
Image generation191 benchmarksArena ELO Text to Image, Arena ELO Text to Image - 3d Modeling, Arena ELO Text to Image - Art…
Vision265 benchmarksVals MMMU Pro, EQ-Bench 3, MMMU-Pro…
All 265 vision benchmarks
Safety101 benchmarksHELM AIR-Bench: AIRBench 2024 - #1: Confidentiality, HELM AIR-Bench: AIRBench 2024 - #10: Depicting Violence, HELM AIR-Bench: AIRBench 2024 - #11: Weapon Usage & Development…
Long context130 benchmarksGDM-MRCRv2: 8 needle, GDM-MRCRv2: 8 needle 128k, GDM-MRCRv2: 8 needle 16k…
OCR42 benchmarksArena ELO Vision - Ocr, OCRBench, OlmOCR Bench…
Embedding326 benchmarksRTEB(deu, beta), RTEB(fra, beta), RTEB(Law, beta)…
Video generation122 benchmarksArena ELO Text to Video, Arena ELO Image to Video, Design Arena Elo (models/imagetovideo)…
Speech560 benchmarksHELM AIR-Bench: AIRBench 2024 - #14: Hate Speech (Inciting/Promoting/Expressing hatred), Gbaeval Audio score, MMAU…
Rerank25 benchmarksBuiltBench(eng) - Reranking, VN-MTEB (vie, v1) - Reranking, Agentset Reranker Elo…
Other1,015 benchmarksDubesor Total, Dubesor STEM, Dubesor Utility…

Most measured

The benchmarks with results for the most models. A benchmark with variants shows once, with its variants listed under it.