llmboard.aiAI model intelligence
Home

Model Rankings

OverallOpen ModelsAgentCodingReasoningMathKnowledgeInstruction FollowingTextVision
Image GenerationImage Editing
Video GenerationImage to VideoVideo Editing
Text to SpeechSpeech to Text
Embeddings

Efficiency

Chat Token PricingImage PricingVideo PricingAudio Pricing
Chat Speed & LatencyProvider Reliability

Benchmarks

GPQAMMLU-ProAIME 2025SWE-Bench VerifiedMMLUHumanity's Last ExamLiveCodeBenchMATHHumanEvalMMMU-Pro
All Benchmarks

Tools

Model DirectoryCompare Models

Scoring & Data

Scoring & Data
393 models668 benchmarks

Leaderboard Center

Overall RankingCodingCore BenchmarksPrice & ValueRuntime Performance

Modalities

All ModelsImage GenerationImage EditingVideo GenerationImage-to-VideoVideo EditingText-to-SpeechSpeech-to-TextEmbeddings

Data & Methods

Scoring MethodAll BenchmarksReasoningMath

Vendors

All VendorsOpenAIAnthropicGoogle
llmboard.aiCopyright 2026 llmboard.ai

legal benchmark

MMLU Leaderboard

Massive Multitask Language Understanding benchmark testing knowledge across 57 diverse subjects including STEM, humanities, social sciences, and professional domains

Updated Aug 17, 2026

Models100
Model coverage101
MetricScore
EvidenceB

On this page

  • Ranking
  • Highlights
  • Distribution
  • Top models
  • About
  • FAQ

MMLU Ranking

Higher score ranks better on this benchmark.

30 of 100 rows
Columns

Show columns

Sort by
Rank
Model
Score
Percentile
Participants
Evidence
Evaluated
Rank01ModelOPGPT-5OpenAIScore92.5%Percentile100.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank02ModelOPo1OpenAIScore91.8%Percentile99.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank03ModelOPGPT-4.5OpenAIScore90.8%Percentile98.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank04ModelOPo1-previewOpenAIScore90.8%Percentile97.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank05ModelACQwen3 VL 235B A22B ThinkingAlibaba Cloud / Qwen TeamScore90.6%Percentile96.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank06ModelSASarvam-105BSarvam AIScore90.6%Percentile95.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank07ModelANClaude 3.5 SonnetAnthropicScore90.4%Percentile94.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank08ModelANClaude 3.5 SonnetAnthropicScore90.4%Percentile93.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank09ModelOPGPT-4.1OpenAIScore90.2%Percentile92.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank10ModelMAKimi K2 0905Moonshot AIScore90.2%Percentile91.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank11ModelOPGPT OSS 120BOpenAIScore90.0%Percentile90.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank12ModelMELongCat-Flash-ChatMeituanScore89.7%Percentile89.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank13ModelMAKimi K2 InstructMoonshot AIScore89.5%Percentile88.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank14ModelMAKimi K2-Instruct-0905Moonshot AIScore89.5%Percentile87.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank15ModelXIMiMo-V2.5-ProXiaomiScore89.4%Percentile86.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank16ModelACQwen3 VL 235B A22B InstructAlibaba Cloud / Qwen TeamScore88.8%Percentile85.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank17ModelOPGPT-4oOpenAIScore88.7%Percentile84.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank18ModelACQwen3 VL 32B ThinkingAlibaba Cloud / Qwen TeamScore88.7%Percentile83.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank19ModelDEDeepSeek-V3DeepSeekScore88.5%Percentile82.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank20ModelACQwen3 235B A22BAlibaba Cloud / Qwen TeamScore87.8%Percentile81.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank21ModelMAKimi K2 BaseMoonshot AIScore87.8%Percentile80.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank22ModelACQwen3 VL 30B A3B ThinkingAlibaba Cloud / Qwen TeamScore87.6%Percentile79.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank23ModelOPGPT-4.1 miniOpenAIScore87.5%Percentile78.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank24ModelXAGrok-2xAIScore87.5%Percentile77.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank25ModelMAKimi-k1.5Moonshot AIScore87.4%Percentile76.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank26ModelMELlama 3.1 405B InstructMetaScore87.3%Percentile75.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank27ModelOPo3-miniOpenAIScore86.9%Percentile74.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank28ModelANClaude 3 OpusAnthropicScore86.8%Percentile73.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank29ModelOPGPT-4 TurboOpenAIScore86.5%Percentile72.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank30ModelOPGPT-4OpenAIScore86.4%Percentile71.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank31ModelACQwen3 VL 32B InstructAlibaba Cloud / Qwen TeamScore86.4%Percentile70.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank32ModelXAGrok-2 minixAIScore86.2%Percentile69.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank33ModelMELlama 3.2 90B InstructMetaScore86.0%Percentile68.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank34ModelMELlama 3.3 70B InstructMetaScore86.0%Percentile67.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank35ModelGOGemini 1.5 ProGoogleScore85.9%Percentile66.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank36ModelAMNova ProAmazonScore85.9%Percentile65.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank37ModelOPGPT-4oOpenAIScore85.7%Percentile64.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank38ModelMELongCat-Flash-LiteMeituanScore85.5%Percentile63.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank39ModelMELlama 4 MaverickMetaScore85.5%Percentile62.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank40ModelOPGPT OSS 20BOpenAIScore85.3%Percentile61.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank41ModelOPo1-miniOpenAIScore85.2%Percentile60.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank42ModelACQwen3 VL 8B ThinkingAlibaba Cloud / Qwen TeamScore85.2%Percentile59.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank43ModelSASarvam-30BSarvam AIScore85.1%Percentile58.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank44ModelACQwen3 VL 30B A3B InstructAlibaba Cloud / Qwen TeamScore85.0%Percentile57.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank45ModelMIPhi 4MicrosoftScore84.8%Percentile56.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank46ModelMAMistral Large 2Mistral AIScore84.0%Percentile55.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank47ModelMELlama 3.1 70B InstructMetaScore83.6%Percentile54.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank48ModelACQwen2.5 32B InstructAlibaba Cloud / Qwen TeamScore83.3%Percentile53.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank49ModelACQwen2 72B InstructAlibaba Cloud / Qwen TeamScore82.3%Percentile52.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank50ModelOPGPT-4o miniOpenAIScore82.0%Percentile51.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank51ModelACQwen3 VL 4B ThinkingAlibaba Cloud / Qwen TeamScore81.5%Percentile50.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank52ModelXAGrok-1.5xAIScore81.3%Percentile49.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank53ModelALJamba 1.5 LargeAI21 LabsScore81.2%Percentile48.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank54ModelMAMistral Small 3.1 24B BaseMistral AIScore81.0%Percentile47.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank55ModelMAMistral Small 3 24B BaseMistral AIScore80.7%Percentile46.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank56ModelACQwen3 VL 8B InstructAlibaba Cloud / Qwen TeamScore80.7%Percentile45.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank57ModelMAMistral Small 3.1 24B InstructMistral AIScore80.6%Percentile44.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank58ModelMAMistral Small 3.2 24B InstructMistral AIScore80.5%Percentile43.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank59ModelAMNova LiteAmazonScore80.5%Percentile42.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank60ModelDEDeepSeek-V2.5DeepSeekScore80.4%Percentile41.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank61ModelNVLlama 3.1 Nemotron 70B InstructNVIDIAScore80.2%Percentile40.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank62ModelOPGPT-4.1 nanoOpenAIScore80.1%Percentile39.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank63ModelACQwen2.5 14B InstructAlibaba Cloud / Qwen TeamScore79.7%Percentile38.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank64ModelMELlama 4 ScoutMetaScore79.6%Percentile37.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank65ModelMAMinistral 3 (14B Base 2512)Mistral AIScore79.4%Percentile36.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank66ModelNRHermes 3 70BNous ResearchScore79.1%Percentile35.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank67ModelANClaude 3 SonnetAnthropicScore79.0%Percentile34.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank68ModelGOGemini 1.5 FlashGoogleScore78.9%Percentile33.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank69ModelMIPhi-3.5-MoE-instructMicrosoftScore78.9%Percentile32.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank70ModelACQwen2.5 VL 32B InstructAlibaba Cloud / Qwen TeamScore78.4%Percentile31.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank71ModelAMNova MicroAmazonScore77.6%Percentile30.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank72ModelACQwen3 VL 4B InstructAlibaba Cloud / Qwen TeamScore77.2%Percentile29.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank73ModelMAMinistral 3 (8B Base 2512)Mistral AIScore76.1%Percentile28.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank74ModelCOCommand R+CohereScore75.7%Percentile27.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank75ModelANClaude 3 HaikuAnthropicScore75.2%Percentile26.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank76ModelGOGemma 2 27BGoogleScore75.2%Percentile25.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank77ModelACQwen2.5-Coder 32B InstructAlibaba Cloud / Qwen TeamScore75.1%Percentile24.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank78ModelMELlama 3.2 11B InstructMetaScore73.0%Percentile23.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank79ModelGOGemini 1.0 ProGoogleScore71.8%Percentile22.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank80ModelGOGemma 2 9BGoogleScore71.3%Percentile21.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank81ModelMAMinistral 3 (3B Base 2512)Mistral AIScore70.7%Percentile20.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank82ModelACQwen2 7B InstructAlibaba Cloud / Qwen TeamScore70.5%Percentile19.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank83ModelOPGPT-3.5 TurboOpenAIScore69.8%Percentile18.0%Participants101EvidenceBEvaluatedAug 17, 2026
Rank84ModelALJamba 1.5 MiniAI21 LabsScore69.7%Percentile17.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank85ModelMELlama 3.1 8B InstructMetaScore69.4%Percentile16.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank86ModelMAPixtral-12BMistral AIScore69.2%Percentile15.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank87ModelMIPhi-3.5-mini-instructMicrosoftScore69.0%Percentile14.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank88ModelMAMistral NeMo InstructMistral AIScore68.0%Percentile13.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank89ModelACQwen2.5-Coder 7B InstructAlibaba Cloud / Qwen TeamScore67.6%Percentile12.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank90ModelMIPhi 4 MiniMicrosoftScore67.3%Percentile11.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank91ModelIBGranite 3.3 8B InstructIBMScore65.5%Percentile10.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank92ModelMAMinistral 8B InstructMistral AIScore65.0%Percentile9.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank93ModelGOGemma 3n E4B InstructedGoogleScore64.9%Percentile8.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank94ModelGOGemma 3n E4B Instructed LiteRT PreviewGoogleScore64.9%Percentile7.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank95ModelIBGranite 3.3 8B BaseIBMScore63.9%Percentile6.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank96ModelMELlama 3.2 3B InstructMetaScore63.4%Percentile5.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank97ModelIBIBM Granite 4.0 Tiny PreviewIBMScore60.4%Percentile4.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank98ModelGOGemma 3n E2B InstructedGoogleScore60.1%Percentile3.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank99ModelGOGemma 3n E2B Instructed LiteRT (Preview)GoogleScore60.1%Percentile2.0%Participants101EvidenceCEvaluatedAug 17, 2026
Rank100ModelCONorth Micro Vision InstructCohereScore50.4%Percentile1.0%Participants101EvidenceCEvaluatedAug 17, 2026

MMLU Highlights

The leading models and scores on this benchmark.

Rank #1GPT-592.5%Rank #2o191.8%Rank #3GPT-4.590.8%Rank #4o1-preview90.8%

MMLU Score Distribution

A closer view of the leading scores on this benchmark.

MMLU

The Top AI Models for MMLU

The first five results on this benchmark, with official price and output speed added where the model identity can be matched.

Ranking basisThis mmlu AI model leaderboard uses descending score in the benchmark's original unit. The leaderboard ranking keeps matched price and speed data separate from benchmark evidence.

  1. 01
    OP
    GPT-5OpenAI
    Score
    92.5%
    Price
    $1.3 input / $10 output per 1M tokens
    Speed
    Up to 100 tok/s via OpenAI

    Strengths

    • Ranks #1 of 101 compared models
    • 100th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures MMLU, not total model capability
  2. 02
    OP
    o1OpenAI
    Score
    91.8%
    Price
    $15 input / $60 output per 1M tokens
    Speed
    Up to 66 tok/s via OpenAI

    Strengths

    • Ranks #2 of 101 compared models
    • 99th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures MMLU, not total model capability
  3. 03
    OP
    GPT-4.5OpenAI
    Score
    90.8%
    Speed
    Up to 50 tok/s via OpenAI

    Strengths

    • Ranks #3 of 101 compared models
    • 98th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures MMLU, not total model capability
  4. 04
    OP
    o1-previewOpenAI
    Score
    90.8%
    Price
    $15 input / $60 output per 1M tokens
    Speed
    Up to 66 tok/s via OpenAI

    Strengths

    • Ranks #4 of 101 compared models
    • 97th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures MMLU, not total model capability
  5. 05
    AC
    Qwen3 VL 235B A22B ThinkingAlibaba Cloud / Qwen Team
    Score
    90.6%

    Strengths

    • Ranks #5 of 101 compared models
    • 96th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures MMLU, not total model capability

Selection summary

Best AI Models for MMLU

GPT-5 currently leads MMLU with 92.5%. It is the top model on this specific benchmark, while the best LLM for the broader task should also be checked against other benchmarks, price and runtime.

Use this leaderboard with the supporting benchmark results and coverage details above. A leaderboard position summarizes the selected ranking signal; it does not replace workload-specific testing.

Benchmark rank #1GPT-592.5% · $1.3 input / $10 output per 1M tokensBenchmark rank #2o191.8% · $15 input / $60 output per 1M tokensBenchmark rank #3GPT-4.590.8% · Up to 50 tok/s via OpenAI

What is MMLU?

What MMLU measures and how its scores work.

Massive Multitask Language Understanding benchmark testing knowledge across 57 diverse subjects including STEM, humanities, social sciences, and professional domains

Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.

Family
MMLU
Modality
text
Primary category
legal
Score direction
higher
LLMBoard eligible
Yes
Evaluation key
mmlu|llm-stats-current

Benchmark scores retain their original unit. Overall score eligibility is shown separately.

FAQ

Common questions about MMLU.

Which model scores highest on MMLU?

GPT-5 is currently ranked first with 92.5%.

What does MMLU measure?

Massive Multitask Language Understanding benchmark testing knowledge across 57 diverse subjects including STEM, humanities, social sciences, and professional domains

Is a higher score better?

Yes. Higher values rank better for this benchmark.

How many models are compared?

100 model results are currently shown.

Does this benchmark affect the overall score?

Yes. This benchmark can contribute to the current LLMBoard capability score.