Benchmark Registry

Humanity's Last Exam

Humanity's Last Exam results reported for 68 models across 7 versions. Updated 2026-10-08.

Report an issue

Versions

7 versions · Metric: Humanity's Last Exam Accuracy

VersionReleasedModelsResults
Humanity's Last Exam No tools — GPT-5.2 evaluationConfiguration: No toolsDecember 11, 202512
Humanity's Last Exam Search and Python — GPT-5.2 evaluationConfiguration: Search and PythonDecember 11, 202511
Humanity's Last Exam Full set — Scale AI evaluationLatestApril 3, 20252735
Humanity's Last Exam Text only — Mercor single-shotConfiguration: Mercor single-shot agentApril 3, 20253947
Humanity's Last Exam Text only — Scale AI evaluationApril 3, 20252634
Humanity's Last Exam Full set — No toolsConfiguration: No toolsJanuary 24, 20252946
Humanity's Last Exam Full set — With toolsConfiguration: With toolsJanuary 24, 20252545

Recently reported results: Humanity's Last Exam Full set — Scale AI evaluation

Latest version: Humanity's Last Exam Full set — Scale AI evaluation

ModelScoreSource
OpenAI o4-mini (high)18.1%labs.scale.comArchived copyHumanity's Last Exam AccuracyReported Benchmark leaderboardScale AIIndependentRecord
GPT-5.227.8%labs.scale.comArchived copyHumanity's Last Exam AccuracyReported Benchmark leaderboardScale AIIndependentRecord
GPT-4.5 Preview5.4%labs.scale.comArchived copyHumanity's Last Exam AccuracyReported Benchmark leaderboardScale AIIndependentRecord
Claude Sonnet 4.5 (thinking)13.7%labs.scale.comArchived copyHumanity's Last Exam AccuracyReported Benchmark leaderboardScale AIIndependentRecord
Claude Sonnet 4.57.5%labs.scale.comArchived copyHumanity's Last Exam AccuracyReported Benchmark leaderboardScale AIIndependentRecord