Reasoning and knowledge
Humanity's Last Exam
4 models with a published score, best first. Higher is better. The record has been broken 4 times since Feb 2, 2025.
67.7%Claude Opus 5.5Rank 1, published Sep 22, 2026
Leaderboard
Each model's latest published score. Every score links to its source.
| # | Model | HLE | Date | Note | Source |
|---|---|---|---|---|---|
| 1 | A Claude Opus 5.5RecordAnthropic | 67.7% | Sep 22, 2026 | with tools, lab-reported | anthropic.com |
| 2 | O GPT-6 AstraOpenAI | 54.8% | Oct 19, 2026 | Scale board | benchlm.ai |
| 3 | A Claude Fable 5.1Anthropic | 46.5% | Oct 19, 2026 | Scale board | benchlm.ai |
| 4 | A Claude Sonnet 5.5Anthropic | 46.1% | Oct 19, 2026 | Scale board | benchlm.ai |
Record history
The best score ever published on Humanity's Last Exam, and who held it. Each step is a new record.
| Date | Record holder | HLE | Gain | Note |
|---|---|---|---|---|
| Sep 22, 2026 | Claude Opus 5.5Current | 67.7% | +3.0 | with tools, lab-reported |
| Apr 7, 2026 | Claude Mythos Preview | 64.7% | +16.3 | unverified |
| Feb 15, 2026 | Gemini Deep Think v2 | 48.4% | +4.0 | without tools, approximate date |
| Jul 9, 2025 | Grok 4 Heavy | 44.4% | +17.8 | with tools, xAI-reported |
| Feb 2, 2025 | OpenAI Deep Research | 26.6% | First | with browsing, lab-reported |
Method
Scale's public board, with lab reports noted per score. Settings vary. Higher is better. Board source: scale.com.
We collect the scores the board and the labs publish and run none ourselves. Ranks use each model's latest reading. 8 readings stored for this board, the newest from Oct 19, 2026. The record history includes older models that no longer have a page here.