Humanity's Last Exam snapshot: result set first checked 22 Oct 2026
Permanent, dated copy. Leaderboard date not shown by the source. Live leaderboard →
Download CSV Every row carries the source date, check dates and source link.
Humanity's Last Exam: Scale AI / CAIS main leaderboard, as ChatJudge copies it
In every ChatJudge check from 22 Oct 2026 to 17 Nov 2026 (the source shows no leaderboard date), the highest reported score on Humanity's Last Exam was GPT 6 Astra (OpenAI) at 54.80%, per Scale AI and Center for AI Safety.
What changed: This is the only Humanity's Last Exam result set ChatJudge has archived: the rows were identical at every check from 22 Oct 2026 to 17 Nov 2026.
- Question set
- All public HLE questions (2,500, including questions with images), per Scale’s methodology
- Settings
- Temperature 0.0 where configurable; answers graded by an o3-mini judge (Scale’s methodology)
- Tools or search
- Not stated by source
- Source date
- Leaderboard date not shown by the source
- Checked by ChatJudge
- 22 Oct 2026 to 17 Nov 2026; the rows were identical at every check
- Rows
- Top 20 of 55 models listed. Model names as listed by Scale. Two models have no published calibration error, so they appear in the table and ranked view only.
| Rank | Model | Provider | Accuracy | Calibration error (lower is better) | ± |
|---|---|---|---|---|---|
| 1 | GPT 6 Astra | OpenAI | 54.80% | 39% | ±1.94 |
| 2 | Fable 5.1 (xhigh) | Anthropic | 46.50% | 20% | ±2.00 |
| 3 | gemini-3.1-pro-preview (thinking high) | 46.44% | 51% | ±1.96 | |
| 4 | Sonnet 5.5 (xhigh) | Anthropic | 46.12% | – | ±1.95 |
| 5 | Gemini 3.8 Flash | 44.52% | 51% | ±1.96 | |
| 6 | gpt-5.4-pro-2026-03-05 | OpenAI | 44.32% | 38% | ±1.95 |
| 7 | Muse Spark | Meta | 40.56% | 50% | ±1.92 |
| 8 | gemini-3-pro-preview | 37.52% | 57% | ±1.90 | |
| 9 | gpt-5.4-2026-03-05 (xhigh thinking) | OpenAI | 36.24% | 42% | ±1.88 |
| 10 | claude-opus-4-7 | Anthropic | 36.20% | 47% | ±1.88 |
| 11 | claude-opus-4-6-thinking-max | Anthropic | 34.44% | 46% | ±1.86 |
| 12 | gpt-5-pro-2025-10-06 | OpenAI | 31.64% | 49% | ±1.82 |
| 13 | Claude Haiku 5.5 | Anthropic | 30.12% | – | ±1.83 |
| 14 | gpt-5.2-2025-12-11 | OpenAI | 27.80% | 45% | ±1.76 |
| 15 | gpt-5-2025-08-07 | OpenAI | 25.32% | 50% | ±1.70 |
| 16 | claude-opus-4-5-20251101-thinking | Anthropic | 25.20% | 55% | ±1.70 |
| 17 | kimi-k2.5 | Moonshot | 24.37% | 67% | ±1.81 |
| 18 | gpt-5.1-thinking | OpenAI | 23.68% | 55% | ±1.67 |
| 19 | gemini-2.5-pro-preview-06-05 | 21.64% | 72% | ±1.61 | |
| 20 | o3 (high) (April 2025) | OpenAI | 20.32% | 34% | ±1.58 |
Humanity's Last Exam is by the Center for AI Safety (CAIS) and Scale AI; the leaderboard is run and published by Scale AI. ChatJudge copies the published numbers and does not run the benchmark. Scores from different views or benchmarks aren’t directly comparable.
