Benchmarks · reasoning
Humanity's Last Exam (text only)
Scale's run of Humanity's Last Exam on its text-only subset (the questions without images): expert-written frontier questions across disciplines, accuracy with a 95% confidence interval. Models evaluated after HLE became public carry a contamination warning.
46 models · 49 results · data as of 2026-10-29 · scale.com ↗
Leaderboard
GPT-6 Astra max
All results
49 rows| # | Model | Effort | Score |
|---|---|---|---|
| 1 | GPT-6 Astra | max | 54.2% ±2.1 |
| 2 | Gemini 3.1 Pro | high | 47.3% ±2.1 |
| 3 | Fable 5.1 | xhigh | 46.8% ±2.1 |
| 4 | Sonnet 5.5 | xhigh | 46.6% ±2.1 |
| 5 | GPT-5.4 Pro | xhigh | 45.3% ±2.1 |
| 6 | Gemini 3.8 Flash | max | 45.0% ±2.1 |
| 7 | Muse Spark | 40.9% ±2.1 | |
| 8 | Gemini 3 Pro | high | 37.7% ±2.0 |
| 9 | GPT-5.4 | xhigh | 36.5% ±2.0 |
| 10 | Opus 4.6 | max | 36.2% ±2.0 |
| 11 | GPT-5 Pro | high | 33.3% ±2.0 |
| 12 | Haiku 5.5 | max | 30.4% ±2.0 |
| 13 | GPT-5.2 | xhigh | 28.5% ±1.9 |
| 14 | Opus 4.5 | high | 26.3% ±1.9 |
| 15 | GPT-5.2025 | high | 26.3% ±1.9 |
| 16 | GPT-5.1 | high | 24.6% ±1.8 |
| 17 | Gemini 2.5 Pro | high | 22.1% ±1.8 |
| 18 | o3 | high | 20.6% ±1.7 |
| 19 | o3 | medium | 19.8% ±1.7 |
| 20 | GPT-5 Mini | high | 19.7% ±1.7 |
| 21 | o4 mini | high | 18.9% ±1.6 |
| 22 | gpt-oss-120b | high | 15.5% ±1.5 |
| 23 | Qwen 3 235B-A22B | xhigh | 15.4% ±1.5 |
| 24 | o4 mini | medium | 14.5% ±1.5 |
| 25 | Sonnet 4.5 | high | 14.1% ±1.5 |
| 26 | DeepSeek R1.0528 | 14.0% ±1.5 | |
| 27 | o3 mini | high | 13.4% ±1.4 |
| 28 | DeepSeek V3.1 | none | 12.9% ±1.4 |
| 29 | Gemini 2.5 Flash | medium | 12.6% ±1.4 |
| 30 | Opus 4.1 | high | 11.3% ±1.3 |