Skip to content

AI benchmark index

AI Benchmark Index catalogs 264 AI benchmarks from primary sources.

Results

264 benchmarks

  • MultimodalOnline

    MLVU

    Tests long-video understanding through multiple-choice and free-form tasks requiring local and global comprehension.

    Task count
    9

  • MultimodalOnline

    MM-Vet

    Measures how well multimodal models integrate recognition, OCR, knowledge, language, spatial, and mathematical capabilities.

  • MultimodalOnline

    MMBench

    Measures broad multimodal model capabilities with multiple-choice questions spanning 20 ability dimensions.

  • MultimodalOnline

    MME

    Measures perception and cognition in multimodal language models across 14 manually designed subtasks.

  • ReasoningOnline

    MMLU

    Tests broad academic knowledge and reasoning with multiple-choice questions drawn from 57 subjects.

    Task count
    15,908

  • ScienceOnline

    MMLU Medical Subjects

    Evaluates factual knowledge and problem solving across the medical and biomedical subject sets in MMLU.

  • ReasoningOnline

    MMLU-Pro

    Evaluates broad knowledge and reasoning with harder multiple-choice questions, expanded answer choices, and reduced ambiguity relative to MMLU.

    Task count
    12,032

  • MultimodalOnline

    MMMU

    Measures expert-level multimodal reasoning across college disciplines, subjects, and heterogeneous image types.

    Task count
    11,500

  • MultimodalOnline

    MMMU-Pro

    Measures robust expert-level multimodal reasoning after filtering text-solvable items and strengthening answer choices.

    Task count
    1,730

  • MultimodalOnline

    MMStar

    Measures vision-essential multimodal capabilities using human-selected samples filtered to resist text-only solving and leakage.

    Task count
    1,500

  • RetrievalOnline

    MMTEB

    Measures multilingual text embeddings across more than 500 quality-controlled tasks covering over 250 languages.

    Task count
    500

  • AgentsOnline

    Mobile-Bench

    Measures mobile agents on instructions that require perception, planning, and actions in Android applications.

  • ScienceOnline

    Mol-Instructions

    Evaluates instruction-following across molecule description, property prediction, reaction, and biomolecular tasks.

  • RetrievalOnline

    MS MARCO

    Evaluates machine reading, answer generation, and passage or document ranking using questions sampled from Bing search.

    Task count
    1,010,916

  • ReasoningOnline

    MT-Bench

    Tests conversational quality on challenging multi-turn, open-ended questions scored by model judges.

    Task count
    80

  • RetrievalOnline

    MTEB

    Measures text embeddings across classification, clustering, retrieval, reranking, similarity, and summarization tasks.

    Task count
    58

  • MemoryOnline

    MultiFieldQA

    Tests single-document and multi-document question answering over long English and Chinese contexts within LongBench.

  • MultimodalOnline

    MVBench

    Measures multi-modal video understanding across 20 temporal and perceptual tasks.

    Task count
    20

  • ReasoningOnline

    Natural Questions

    Tests question answering on real Google queries paired with Wikipedia pages and short or long answer annotations.

    Task count
    323,044

  • MultimodalOnline

    NExT-QA

    Evaluates causal and temporal reasoning in video question answering rather than simple visual description.

  • MultimodalOnline

    OCRBench

    Measures OCR in multimodal models across text recognition, scene-text VQA, document VQA, extraction, and handwritten math.

    Task count
    1,000

  • MultimodalOnline

    OCRBench v2

    Measures bilingual visual-text localization, recognition, extraction, and reasoning across diverse real-world scenarios.

    Task count
    10,000

  • ScienceOnline

    OlympiadBench

    Tests bilingual, multimodal problem solving on olympiad-level mathematics and physics questions.

    Task count
    8,476

  • AgentsOnline

    OSWorld-Verified

    Measures multimodal agents on reproducible, open-ended tasks in real desktop applications; the verified release repairs the original task set.

    Task count
    369

Cities in Global

Categories in Global

Questions

Which cities in Global have the most benchmarks?

The cities in Global with the most benchmarks are Online (264 benchmarks).

What kinds of benchmarks are in Global?

Categories in Global: Multimodal, Reasoning, Agents, Safety, and Science.

Published by . How we check each benchmark.