# Humanity's Last Exam (HLE)

About 2,500 extremely hard questions written by experts in over 100 subjects, from classics to quantum physics, meant to sit at the edge of human knowledge. It is one of the hardest general-knowledge exams for AI.

Category: reasoning. Type: dataset. Status: active.
Metric: Share of questions answered correctly (%) (higher is better).
Source: Epoch AI (Benchmarking Hub) — https://epoch.ai/data/benchmark_data.zip, results as of 2026-11-17, retrieved 2026-11-17, license: CC-BY-4.0.

| Rank | Model | Organization | Accuracy (%) | Calibration error (%) | Release date |
|---|---|---|---|---|---|
| 1 | GPT 6 Astra | OpenAI | 54.8 | 39.0 | 2026-09-03 |
| 2 | Fable 5.1 (xhigh) | Anthropic | 46.5 | 20.0 | 2026-09-01 |
| 3 | gemini-3.1-pro-preview (thinking high) | Google DeepMind | 46.4 | 51.0 | 2026-02-19 |
| 4 | Sonnet 5.5 (xhigh) | Anthropic | 46.1 |  | 2026-09-28 |
| 5 | Gemini 3.8 Flash | Google DeepMind | 44.5 | 51.0 | 2026-09-02 |
| 6 | gpt-5.4-pro-2026-03-05 | OpenAI | 44.3 | 38.0 | 2026-03-05 |
| 7 | Muse Spark | Meta AI | 40.6 | 50.0 | 2026-04-08 |
| 8 | gemini-3-pro-preview | Google DeepMind | 37.5 | 57.0 | 2025-11-18 |
| 9 | gpt-5.4-2026-03-05 (xhigh thinking) | OpenAI | 36.2 | 42.0 | 2026-03-05 |
| 10 | claude-opus-4-7 | Anthropic | 36.2 | 47.0 | 2026-04-16 |
| 11 | claude-opus-4-6-thinking-max | Anthropic | 34.4 | 46.0 | 2026-02-05 |
| 12 | gpt-5-pro-2025-10-06 | OpenAI | 31.6 | 49.0 | 2025-10-07 |
| 13 | Claude Haiku 5.5 | Anthropic | 30.1 |  | 2026-10-07 |
| 14 | gpt-5.2-2025-12-11 | OpenAI | 27.8 | 45.0 | 2025-12-11 |
| 15 | gpt-5-2025-08-07 | OpenAI | 25.3 | 50.0 | 2025-08-07 |
| 16 | gpt-5-2025-08-07 | OpenAI | 25.3 | 50.0 | 2025-08-07 |
| 17 | claude-opus-4-5-20251101-thinking | Anthropic | 25.2 | 55.0 | 2025-11-24 |
| 18 | kimi-k2.5 | Moonshot | 24.4 | 67.0 | 2026-01-27 |
| 19 | gpt-5.1-thinking | OpenAI | 23.7 | 55.0 | 2025-11-13 |
| 20 | gemini-2.5-pro-preview-06-05 | Google DeepMind | 21.6 | 72.0 | 2025-06-05 |
| 21 | o3 (high) (April 2025) | OpenAI | 20.3 | 34.0 | 2025-04-16 |
| 22 | gpt-5-mini-2025-08-07 | OpenAI | 19.4 | 65.0 | 2025-08-07 |
| 23 | o3 (medium) (April 2025) | OpenAI | 19.2 | 39.0 | 2025-04-16 |
| 24 | claude-opus-4-6 (Non-Thinking) | Anthropic | 19.0 | 44.0 | 2026-02-05 |
| 25 | Gemini 2.5 Pro Experimental (March 2025) | Google DeepMind | 18.2 | 71.0 | 2025-03-25 |

Notes: Externally reported results mirrored by Epoch AI (hle_external.csv); accuracy converted to % (1 decimal), calibration error copied as stored. Audits have found wrong reference answers in part of the set.
