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

math benchmark

Humanity's Last Exam Leaderboard

Humanity's Last Exam (HLE) is a multi-modal academic benchmark with 2,500 questions across mathematics, humanities, and natural sciences, designed to test LLM capabilities at the frontier of human knowledge with unambiguous, verifiable solutions

Updated Aug 17, 2026

Models99
Model coverage99
MetricScore
EvidenceB

On this page

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

Humanity's Last Exam Ranking

Higher score ranks better on this benchmark.

30 of 99 rows
Columns

Show columns

Sort by
Rank
Model
Score
Percentile
Participants
Evidence
Evaluated
Rank01ModelANClaude Mythos PreviewAnthropicScore64.7%Percentile100.0%Participants99EvidenceCEvaluatedAug 17, 2026
Rank02ModelANClaude Opus 5AnthropicScore64.7%Percentile99.0%Participants99EvidenceCEvaluatedAug 17, 2026
Rank03ModelANClaude Fable 5AnthropicScore64.5%Percentile98.0%Participants99EvidenceCEvaluatedAug 17, 2026
Rank04ModelZAGLM-5.3Zhipu AIScore62.5%Percentile96.9%Participants99EvidenceCEvaluatedAug 17, 2026
Rank05ModelMEMuse Spark 1.1MetaScore62.1%Percentile95.9%Participants99EvidenceCEvaluatedAug 17, 2026
Rank06ModelDEDeepSeek-V4-Pro-0813DeepSeekScore60.0%Percentile94.9%Participants99EvidenceCEvaluatedAug 17, 2026
Rank07ModelMEMuse SparkMetaScore58.4%Percentile93.9%Participants99EvidenceCEvaluatedAug 17, 2026
Rank08ModelANClaude Opus 4.8AnthropicScore57.9%Percentile92.9%Participants99EvidenceCEvaluatedAug 17, 2026
Rank09ModelANClaude Sonnet 5AnthropicScore57.4%Percentile91.8%Participants99EvidenceCEvaluatedAug 17, 2026
Rank10ModelOPGPT-5.5 ProOpenAIScore57.2%Percentile90.8%Participants99EvidenceCEvaluatedAug 17, 2026
Rank11ModelMAKimi K3Moonshot AIScore56.0%Percentile89.8%Participants99EvidenceCEvaluatedAug 17, 2026
Rank12ModelBYSeed 2.1 ProByteDanceScore55.7%Percentile88.8%Participants99EvidenceCEvaluatedAug 17, 2026
Rank13ModelANClaude Opus 4.7AnthropicScore54.7%Percentile87.8%Participants99EvidenceCEvaluatedAug 17, 2026
Rank14ModelZAGLM-5.2Zhipu AIScore54.7%Percentile86.7%Participants99EvidenceCEvaluatedAug 17, 2026
Rank15ModelBYSeed 2.1 TurboByteDanceScore54.6%Percentile85.7%Participants99EvidenceCEvaluatedAug 17, 2026
Rank16ModelANClaude Opus 4.6AnthropicScore53.1%Percentile84.7%Participants99EvidenceCEvaluatedAug 17, 2026
Rank17ModelZAGLM-5.1Zhipu AIScore52.3%Percentile83.7%Participants99EvidenceCEvaluatedAug 17, 2026
Rank18ModelOPGPT-5.5OpenAIScore52.2%Percentile82.7%Participants99EvidenceCEvaluatedAug 17, 2026
Rank19ModelGOGemini 3.1 ProGoogleScore51.4%Percentile81.6%Participants99EvidenceCEvaluatedAug 17, 2026
Rank20ModelMAKimi K2-Thinking-0905Moonshot AIScore51.0%Percentile80.6%Participants99EvidenceCEvaluatedAug 17, 2026
Rank21ModelXAGrok-4 HeavyxAIScore50.7%Percentile79.6%Participants99EvidenceCEvaluatedAug 17, 2026
Rank22ModelMAKimi K2.5Moonshot AIScore50.2%Percentile78.6%Participants99EvidenceCEvaluatedAug 17, 2026
Rank23ModelANClaude Sonnet 4.6AnthropicScore49.0%Percentile77.5%Participants99EvidenceCEvaluatedAug 17, 2026
Rank24ModelACQwen3.5-27BAlibaba Cloud / Qwen TeamScore48.5%Percentile76.5%Participants99EvidenceCEvaluatedAug 17, 2026
Rank25ModelDEDeepSeek-V4-Pro-MaxDeepSeekScore48.2%Percentile75.5%Participants99EvidenceCEvaluatedAug 17, 2026
Rank26ModelACQwen3.5-122B-A10BAlibaba Cloud / Qwen TeamScore47.5%Percentile74.5%Participants99EvidenceCEvaluatedAug 17, 2026
Rank27ModelACQwen3.5-35B-A3BAlibaba Cloud / Qwen TeamScore47.4%Percentile73.5%Participants99EvidenceCEvaluatedAug 17, 2026
Rank28ModelGOGemini 3 ProGoogleScore45.8%Percentile72.5%Participants99EvidenceCEvaluatedAug 17, 2026
Rank29ModelDEDeepSeek-V4-Flash-MaxDeepSeekScore45.1%Percentile71.4%Participants99EvidenceCEvaluatedAug 17, 2026
Rank30ModelACQwen3.8 MaxAlibaba Cloud / Qwen TeamScore43.6%Percentile70.4%Participants99EvidenceCEvaluatedAug 17, 2026
Rank31ModelGOGemini 3 FlashGoogleScore43.5%Percentile69.4%Participants99EvidenceCEvaluatedAug 17, 2026
Rank32ModelZAGLM-4.7Zhipu AIScore42.8%Percentile68.4%Participants99EvidenceCEvaluatedAug 17, 2026
Rank33ModelACQwen3.7 MaxAlibaba Cloud / Qwen TeamScore41.4%Percentile67.3%Participants99EvidenceCEvaluatedAug 17, 2026
Rank34ModelDEDeepSeek-V3.2DeepSeekScore40.8%Percentile66.3%Participants99EvidenceCEvaluatedAug 17, 2026
Rank35ModelDEDeepSeek-V4-Flash-0423DeepSeekScore40.3%Percentile65.3%Participants99EvidenceCEvaluatedAug 17, 2026
Rank36ModelGOGemini 3.5 FlashGoogleScore40.2%Percentile64.3%Participants99EvidenceCEvaluatedAug 17, 2026
Rank37ModelXAGrok-4xAIScore40.0%Percentile63.3%Participants99EvidenceCEvaluatedAug 17, 2026
Rank38ModelOPGPT-5.4OpenAIScore39.8%Percentile62.2%Participants99EvidenceCEvaluatedAug 17, 2026
Rank39ModelBAERNIE 5.0BaiduScore39.0%Percentile61.2%Participants99EvidenceCEvaluatedAug 17, 2026
Rank40ModelNVNemotron 3 Ultra (550B A55B)NVIDIAScore37.4%Percentile60.2%Participants99EvidenceCEvaluatedAug 17, 2026
Rank41ModelOPGPT-5.2 ProOpenAIScore36.6%Percentile59.2%Participants99EvidenceCEvaluatedAug 17, 2026
Rank42ModelMAKimi K2.6Moonshot AIScore36.4%Percentile58.2%Participants99EvidenceCEvaluatedAug 17, 2026
Rank43ModelACQwen3.7-PlusAlibaba Cloud / Qwen TeamScore34.7%Percentile57.1%Participants99EvidenceCEvaluatedAug 17, 2026
Rank44ModelOPGPT-5.2OpenAIScore34.5%Percentile56.1%Participants99EvidenceCEvaluatedAug 17, 2026
Rank45ModelXIMiMo-V2.5-ProXiaomiScore34.0%Percentile55.1%Participants99EvidenceCEvaluatedAug 17, 2026
Rank46ModelTMInkling-SmallThinking Machines LabScore31.6%Percentile54.1%Participants99EvidenceCEvaluatedAug 17, 2026
Rank47ModelACQwen3.8-27BAlibaba Cloud / Qwen TeamScore30.8%Percentile53.1%Participants99EvidenceCEvaluatedAug 17, 2026
Rank48ModelDEDeepSeek-V3.2-SpecialeDeepSeekScore30.6%Percentile52.0%Participants99EvidenceCEvaluatedAug 17, 2026
Rank49ModelACQwen3.6 PlusAlibaba Cloud / Qwen TeamScore28.8%Percentile51.0%Participants99EvidenceCEvaluatedAug 17, 2026
Rank50ModelACQwen3.5-397B-A17BAlibaba Cloud / Qwen TeamScore28.7%Percentile50.0%Participants99EvidenceCEvaluatedAug 17, 2026
Rank51ModelOPGPT-5.4 miniOpenAIScore28.2%Percentile49.0%Participants99EvidenceCEvaluatedAug 17, 2026
Rank52ModelGOGemma 4 31BGoogleScore26.5%Percentile48.0%Participants99EvidenceCEvaluatedAug 17, 2026
Rank53ModelMELongCat-Flash-Thinking-2601MeituanScore25.2%Percentile46.9%Participants99EvidenceCEvaluatedAug 17, 2026
Rank54ModelDEDeepSeek-V3.2 (Thinking)DeepSeekScore25.1%Percentile45.9%Participants99EvidenceCEvaluatedAug 17, 2026
Rank55ModelOPGPT-5OpenAIScore24.8%Percentile44.9%Participants99EvidenceCEvaluatedAug 17, 2026
Rank56ModelOPGPT-5.4 nanoOpenAIScore24.3%Percentile43.9%Participants99EvidenceCEvaluatedAug 17, 2026
Rank57ModelACQwen3.6-27BAlibaba Cloud / Qwen TeamScore24.0%Percentile42.9%Participants99EvidenceCEvaluatedAug 17, 2026
Rank58ModelNVNemotron 3 Super (120B A12B)NVIDIAScore22.8%Percentile41.8%Participants99EvidenceCEvaluatedAug 17, 2026
Rank59ModelXIMiMo-V2-FlashXiaomiScore22.1%Percentile40.8%Participants99EvidenceCEvaluatedAug 17, 2026
Rank60ModelMIMiniMax M2.1MiniMaxScore22.0%Percentile39.8%Participants99EvidenceCEvaluatedAug 17, 2026
Rank61ModelMEMuse Glimmer-30BMetaScore22.0%Percentile38.8%Participants99EvidenceCEvaluatedAug 17, 2026
Rank62ModelGOGemini 2.5 Pro Preview 06-05GoogleScore21.6%Percentile37.8%Participants99EvidenceCEvaluatedAug 17, 2026
Rank63ModelACQwen3.6-35B-A3BAlibaba Cloud / Qwen TeamScore21.4%Percentile36.7%Participants99EvidenceCEvaluatedAug 17, 2026
Rank64ModelXAGrok 4 FastxAIScore20.0%Percentile35.7%Participants99EvidenceCEvaluatedAug 17, 2026
Rank65ModelDEDeepSeek-V3.2-ExpDeepSeekScore19.8%Percentile34.7%Participants99EvidenceCEvaluatedAug 17, 2026
Rank66ModelACQwen3-235B-A22B-Thinking-2507Alibaba Cloud / Qwen TeamScore18.2%Percentile33.7%Participants99EvidenceCEvaluatedAug 17, 2026
Rank67ModelMIMAI-Code-1-FlashMicrosoftScore18.0%Percentile32.6%Participants99EvidenceCEvaluatedAug 17, 2026
Rank68ModelGOGemini 2.5 ProGoogleScore17.8%Percentile31.6%Participants99EvidenceCEvaluatedAug 17, 2026
Rank69ModelDEDeepSeek-R1-0528DeepSeekScore17.7%Percentile30.6%Participants99EvidenceCEvaluatedAug 17, 2026
Rank70ModelGOGemma 4 26B-A4BGoogleScore17.2%Percentile29.6%Participants99EvidenceCEvaluatedAug 17, 2026
Rank71ModelZAGLM-4.6Zhipu AIScore17.2%Percentile28.6%Participants99EvidenceCEvaluatedAug 17, 2026
Rank72ModelOPGPT-5 miniOpenAIScore16.7%Percentile27.6%Participants99EvidenceCEvaluatedAug 17, 2026
Rank73ModelGOGemini 3.1 Flash-LiteGoogleScore16.0%Percentile26.5%Participants99EvidenceCEvaluatedAug 17, 2026
Rank74ModelDEDeepSeek-V3.1DeepSeekScore15.9%Percentile25.5%Participants99EvidenceCEvaluatedAug 17, 2026
Rank75ModelNVNemotron 3 Nano (30B A3B)NVIDIAScore15.5%Percentile24.5%Participants99EvidenceCEvaluatedAug 17, 2026
Rank76ModelOPGPT OSS 120BOpenAIScore14.9%Percentile23.5%Participants99EvidenceCEvaluatedAug 17, 2026
Rank77ModelOPo3OpenAIScore14.7%Percentile22.4%Participants99EvidenceCEvaluatedAug 17, 2026
Rank78ModelOPo4-miniOpenAIScore14.7%Percentile21.4%Participants99EvidenceCEvaluatedAug 17, 2026
Rank79ModelZAGLM-4.5Zhipu AIScore14.4%Percentile20.4%Participants99EvidenceCEvaluatedAug 17, 2026
Rank80ModelZAGLM-4.7-FlashZhipu AIScore14.4%Percentile19.4%Participants99EvidenceCEvaluatedAug 17, 2026
Rank81ModelACQwen3 VL 235B A22B ThinkingAlibaba Cloud / Qwen TeamScore13.6%Percentile18.4%Participants99EvidenceCEvaluatedAug 17, 2026
Rank82ModelMIMiniMax M2MiniMaxScore12.5%Percentile17.4%Participants99EvidenceCEvaluatedAug 17, 2026
Rank83ModelGODiffusionGemma 26B-A4BGoogleScore11.9%Percentile16.3%Participants99EvidenceCEvaluatedAug 17, 2026
Rank84ModelNVNemotron 3.5 Lightning (30B A3B)NVIDIAScore11.7%Percentile15.3%Participants99EvidenceCEvaluatedAug 17, 2026
Rank85ModelSASarvam-105BSarvam AIScore11.2%Percentile14.3%Participants99EvidenceCEvaluatedAug 17, 2026
Rank86ModelGOGemini 2.5 FlashGoogleScore11.0%Percentile13.3%Participants99EvidenceCEvaluatedAug 17, 2026
Rank87ModelOPGPT OSS 20BOpenAIScore10.9%Percentile12.2%Participants99EvidenceCEvaluatedAug 17, 2026
Rank88ModelZAGLM-4.5-AirZhipu AIScore10.6%Percentile11.2%Participants99EvidenceCEvaluatedAug 17, 2026
Rank89ModelMAMagistral MediumMistral AIScore9.0%Percentile10.2%Participants99EvidenceCEvaluatedAug 17, 2026
Rank90ModelOPGPT-5 nanoOpenAIScore8.7%Percentile9.2%Participants99EvidenceCEvaluatedAug 17, 2026
Rank91ModelMIMiniMax M1 80KMiniMaxScore8.4%Percentile8.2%Participants99EvidenceCEvaluatedAug 17, 2026
Rank92ModelMIMiniMax M1 40KMiniMaxScore7.2%Percentile7.1%Participants99EvidenceCEvaluatedAug 17, 2026
Rank93ModelOPGPT-4.1OpenAIScore5.4%Percentile6.1%Participants99EvidenceCEvaluatedAug 17, 2026
Rank94ModelOPGPT-4oOpenAIScore5.3%Percentile5.1%Participants99EvidenceCEvaluatedAug 17, 2026
Rank95ModelGOGemma 4 12BGoogleScore5.2%Percentile4.1%Participants99EvidenceCEvaluatedAug 17, 2026
Rank96ModelGOGemini 2.5 Flash-LiteGoogleScore5.1%Percentile3.1%Participants99EvidenceCEvaluatedAug 17, 2026
Rank97ModelMAKimi K2 InstructMoonshot AIScore4.7%Percentile2.0%Participants99EvidenceCEvaluatedAug 17, 2026
Rank98ModelMAKimi K2-Instruct-0905Moonshot AIScore4.7%Percentile1.0%Participants99EvidenceCEvaluatedAug 17, 2026
Rank99ModelOPGPT-4.1 miniOpenAIScore3.7%Percentile0.0%Participants99EvidenceCEvaluatedAug 17, 2026

Humanity's Last Exam Highlights

The leading models and scores on this benchmark.

Rank #1Claude Mythos Preview64.7%Rank #2Claude Opus 564.7%Rank #3Claude Fable 564.5%Rank #4GLM-5.362.5%

Humanity's Last Exam Score Distribution

A closer view of the leading scores on this benchmark.

Humanity's Last Exam

The Top AI Models for Humanity's Last Exam

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

Ranking basisThis humanity's last exam 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
    AN
    Claude Mythos PreviewAnthropic
    Score
    64.7%

    Strengths

    • Ranks #1 of 99 compared models
    • 100th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures Humanity's Last Exam, not total model capability
  2. 02
    AN
    Claude Opus 5Anthropic
    Score
    64.7%
    Price
    $5.0 input / $25 output per 1M tokens
    Speed
    Up to 3.3 tok/s via Anthropic

    Strengths

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

    Considerations

    • This result measures Humanity's Last Exam, not total model capability
  3. 03
    AN
    Claude Fable 5Anthropic
    Score
    64.5%
    Price
    $10 input / $50 output per 1M tokens
    Speed
    Up to 9.1 tok/s via Anthropic

    Strengths

    • Ranks #3 of 99 compared models
    • 98th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures Humanity's Last Exam, not total model capability
  4. 04
    ZA
    GLM-5.3Zhipu AI
    Score
    62.5%

    Strengths

    • Ranks #4 of 99 compared models
    • 97th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures Humanity's Last Exam, not total model capability
  5. 05
    ME
    Muse Spark 1.1Meta
    Score
    62.1%
    Price
    $1.3 input / $4.3 output per 1M tokens
    Speed
    Up to 6.2 tok/s via Meta Model API

    Strengths

    • Ranks #5 of 99 compared models
    • 96th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures Humanity's Last Exam, not total model capability

Selection summary

Best AI Models for Humanity's Last Exam

Claude Mythos Preview currently leads Humanity's Last Exam with 64.7%. 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 #1Claude Mythos Preview64.7%Benchmark rank #2Claude Opus 564.7% · $5.0 input / $25 output per 1M tokensBenchmark rank #3Claude Fable 564.5% · $10 input / $50 output per 1M tokens

What is Humanity's Last Exam?

What Humanity's Last Exam measures and how its scores work.

Humanity's Last Exam (HLE) is a multi-modal academic benchmark with 2,500 questions across mathematics, humanities, and natural sciences, designed to test LLM capabilities at the frontier of human knowledge with unambiguous, verifiable solutions

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

Family
Humanity's Last Exam
Modality
multimodal
Primary category
math
Score direction
higher
LLMBoard eligible
Yes
Evaluation key
humanity's-last-exam|llm-stats-current

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

FAQ

Common questions about Humanity's Last Exam.

Which model scores highest on Humanity's Last Exam?

Claude Mythos Preview is currently ranked first with 64.7%.

What does Humanity's Last Exam measure?

Humanity's Last Exam (HLE) is a multi-modal academic benchmark with 2,500 questions across mathematics, humanities, and natural sciences, designed to test LLM capabilities at the frontier of human knowledge with unambiguous, verifiable solutions

Is a higher score better?

Yes. Higher values rank better for this benchmark.

How many models are compared?

99 model results are currently shown.

Does this benchmark affect the overall score?

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