evals.report
BenchmarksLabsCompareRun guidesIn the wild
BenchmarksReasoning

Humanity's Last Exam

A broad expert-level academic question-answering benchmark for frontier reasoning systems.

ReasoningaccuracyHigher is better
ScoresAboutRun this benchmark

What is Humanity's Last Exam?

A broad expert-level academic question-answering benchmark for frontier reasoning systems. evals.report tracks reported Humanity's Last Exam scores with the model, source, status, date, and run caveats attached — official leaderboard scores, vendor-reported launches, and clearly labeled community runs.

Top reported Humanity's Last Exam score: Claude Mythos Preview — 56.8% (accuracy).

ModelLabScore↓Source modelStatusDate
Claude Mythos PreviewAnthropic56.8%Claude Mythos PreviewUnverifiedApr 7, 2026Details
Muse Spark 1.1Meta52.2%Muse Spark 1.1VerifiedJul 9, 2026Details
Fugu UltraSakana AI50.0%Fugu UltraVerifiedJun 15, 2026Details
Claude Opus 4.8Anthropic49.8%Claude Opus 4.8VerifiedMay 28, 2026Details
FuguSakana AI47.2%FuguVerifiedJun 15, 2026Details
Gemini 3.1 Pro PreviewGoogle DeepMind45.9%Gemini 3.1 ProOfficialFeb 19, 2026Details
GPT-5.5OpenAI43.56%GPT-5.5OfficialApr 23, 2026Details
Claude Sonnet 5Anthropic43.2%Claude Sonnet 5VerifiedJun 30, 2026Details
Gemini 3.5 FlashGoogle DeepMind42.5%Gemini 3.5 FlashOfficialMay 19, 2026Details
GLM-5.2OpenZ.ai40.5%GLM-5.2VerifiedJun 16, 2026Details
GPT-5.4OpenAI40.28%GPT-5.4OfficialMar 5, 2026Details
Claude Opus 4.7Anthropic39.04%Opus 4.7OfficialApr 16, 2026Details
Gemini 3 ProGoogle DeepMind38.3%Gemini 3 ProOfficialNov 18, 2025Details
Gemini 3 FlashGoogle DeepMind36.6%Gemini 3 FlashOfficialDec 17, 2025Details
DeepSeek V4 FlashOpenDeepSeek34.8%DeepSeek-V4-Flash (Think-Max)VerifiedApr 24, 2026Details
Claude Opus 4.6Anthropic34.2%Opus 4.6OfficialFeb 5, 2026Details
MiMo-V2.5-ProOpenXiaomi33.8%MiMo-V2.5-ProVerifiedApr 22, 2026Details
Grok 4.3xAI33.12%Grok 4.3OfficialApr 17, 2026Details
DeepSeek V4 ProOpenDeepSeek32.4%DeepSeek 4 ProOfficialApr 24, 2026Details
Doubao Seed 2.0 ProByteDance32.4%Seed2.0 Pro (0215)VerifiedFeb 14, 2026Details
Grok 4.2xAI30.2%Grok 4.2OfficialFeb 17, 2026Details
Kimi K2.6OpenMoonshot AI29.9%Kimi K2.6OfficialApr 20, 2026Details
GPT-5.2OpenAI29.9%GPT-5.2OfficialDec 11, 2025Details
InklingOpenThinking Machines Lab29.7%InklingVerifiedJul 15, 2026Details
Qwen3.5-397B-A17BOpenAlibaba / Qwen28.7%Qwen3.5-397B-A17BUnverifiedFeb 16, 2026Details
GPT-5.1OpenAI27.2%GPT-5.1OfficialNov 12, 2025Details
NVIDIA Nemotron 3 UltraOpenNVIDIA26.7%Nemotron-3-Ultra-550B-A55B (BF16)VerifiedJun 4, 2026Details
Claude Opus 4.5Anthropic25.8%Opus 4.5OfficialNov 24, 2025Details
GLM-5.1OpenZ.ai25.63%GLM 5.1OfficialApr 7, 2026Details
GPT-5OpenAI25.32%GPT-5OfficialAug 7, 2025Details
GLM-4.7OpenZ.ai24.8%GLM-4.7VerifiedDec 22, 2025Details
Grok 4xAI24.52%Grok 4OfficialJul 9, 2025Details
Kimi K2 ThinkingOpenMoonshot AI23.9%Kimi K2 ThinkingVerifiedNov 6, 2025Details
Gemini 2.5 ProGoogle DeepMind21.64%Gemini 2.5 ProOfficialMar 25, 2025Details
Claude Sonnet 4.6Anthropic21.07%Sonnet 4.6OfficialFeb 17, 2026Details
GPT-5 miniOpenAI19.4%GPT-5-miniOfficialAug 7, 2025Details
MiniMax M2.5OpenMiniMax19.4%MiniMax-M2.5VerifiedFeb 12, 2026Details
NVIDIA Nemotron 3 Super 120B-A12BOpenNVIDIA18.26%N-3-SuperVerifiedMar 10, 2026Details
MAI-Code-1-FlashMicrosoft AI18%MAI-Code-1-FlashUnverifiedJun 2, 2026Details
DeepSeek V3.1OpenDeepSeek15.9%DeepSeek-V3.1-ThinkingVerifiedAug 21, 2025Details
Claude Sonnet 4.5Anthropic13.7%Claude 4.5 SonnetOfficialSep 29, 2025Details
K-EXAONEOpenLG AI Research13.6%K-EXAONE (Reasoning)UnverifiedJan 12, 2026Details
Gemini 2.5 FlashGoogle DeepMind12.1%Gemini 2.5 FlashOfficialApr 17, 2025Details
DeepSeek R1OpenDeepSeek8.5%DeepSeek-R1*OfficialJan 20, 2025Details
Reka Flash 3OpenReka AI5.1%Reka Flash 3UnverifiedMar 10, 2025Details
Jamba 1.7 LargeOpenAI21 Labs3.8%Jamba 1.7 LargeUnverifiedJul 3, 2025Details
GPT-4oOpenAI2.7%GPT-4oOfficialMay 13, 2024Details

Each row reports the model’s accuracy on Humanity's Last Exam. Click a row for the full run context.