HLE-Diamond: leaderboard
CAIS and Scale AI's refined 1,000-question subset of Humanity's Last Exam, after a year of cleaning: 500 reasoning and 500 expert-knowledge questions, answered closed-book without tools.
Metric: Accuracy (%, no tools). Source: lastexam.ai. Saturation forecast: Around June 2027. 32 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 65.8 | #6 (GPT-6 Astra) |
| 2 | GPT-6.1 Sol (Max) | 61.3 | #10 (GPT-6.1 Sol) |
| 3 | Claude Opus 5.5 (Max) | 60.9 | #5 (Claude Opus 5.5) |
| 4 | GPT-6 Astra (High) | 59.9 | #6 (GPT-6 Astra) |
| 5 | Gemini 4 Argon (High) | 56.9 | #11 (Gemini 4 Argon) |
| 6 | Claude Fable 5.1 (Max) | 55.1 | #14 (Claude Fable 5.1) |
| 7 | Claude Opus 5.5 (High) | 54.6 | #5 (Claude Opus 5.5) |
| 8 | GPT-6.1 Sol (High) | 53.2 | #10 (GPT-6.1 Sol) |
| 9 | Claude Fable 5.1 (High) | 51.6 | #14 (Claude Fable 5.1) |
| 10 | Claude Sonnet 5.5 (Max) | 43.3 | #18 (Claude Sonnet 5.5) |
| 11 | GPT-6 Sol (Max) | 43.2 | #24 (GPT-6 Sol) |
| 12 | GPT-5.6 Sol (Max) | 41 | #29 (GPT-5.6 Sol) |
| 13 | Muse Wonder (Max) | 40.6 | #27 (Muse Wonder) |
| 14 | Claude Opus 5 (Max) | 40.5 | #20 (Claude Opus 5) |
| 15 | Kimi K3.1 Pro (Max) | 38.7 | #28 (Kimi K3.1 Pro) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon.
Interactive version: theaggregate.ai/benchmark?slug=hle-diamond · How It Works · Data refreshed daily, snapshot 2026-11-12.