Humanity's Last Exam
2,500 expert-written questions at the edge of human knowledge.
Agent Index weight
0%
shown for context only
Models tested
43
from the benchmark’s leaderboard and reports
Results
| # | Model | Score | Normalised | Harness | Cost/task | Date |
|---|---|---|---|---|---|---|
| 1 | GPT-6 Astra Supported OpenAI | 54.8% | source | — | — | |
| 2 | Claude Fable 5.1 Supported Anthropic | 46.5% | source | — | — | |
| 3 | Gemini 3.1 Pro Preview Supported Google | 46.4% | source | — | — | |
| 4 | Claude Sonnet 5.5 Estimated Anthropic | 46.1% | source | — | — | |
| 5 | Gemini 3.8 Flash Supported Google | 44.5% | source | — | — | |
| 6 | GPT-5.4 Pro OpenAI | 44.3% | source | — | — | |
| 7 | Muse Spark Meta | 40.6% | source | — | — | |
| 8 | Gemini 3 Pro Preview Supported Google | 37.5% | source | — | — | |
| 9 | GPT-5.4 Supported OpenAI | 36.2% | source | — | — | |
| 10 | Claude Opus 4.7 Supported Anthropic | 36.2% | source | — | — | |
| 11 | Claude Opus 4.6 Supported Anthropic | 34.4% | source | — | — | |
| 12 | GPT-5 Pro OpenAI | 31.6% | source | — | — | |
| 13 | Claude Haiku 5.5 Estimated Anthropic | 30.1% | source | — | — | |
| 14 | GPT-5.2 Supported OpenAI | 27.8% | source | — | — | |
| 15 | GPT-5 Supported OpenAI | 25.3% | source | — | — | |
| 16 | Claude Opus 4.5 (latest) Supported Anthropic | 25.2% | source | — | — | |
| 17 | Kimi K2.5 Supported Moonshot AI | 24.4% | source | — | — | |
| 18 | GPT-5.1 Supported OpenAI | 23.7% | source | — | — | |
| 19 | Gemini 2.5 Pro Supported Google | 21.6% | source | — | — | |
| 20 | o3 Supported OpenAI | 20.3% | source | — | — | |
| 21 | GPT-5 Mini Estimated OpenAI | 19.4% | source | — | — | |
| 22 | o4-mini Estimated OpenAI | 18.1% | source | — | — | |
| 23 | Claude Sonnet 4.5 (latest) Supported Anthropic | 13.7% | source | — | — | |
| 24 | Gemini 2.5 Flash Estimated Google | 12.1% | source | — | — | |
| 25 | Claude Opus 4.1 (latest) Supported Anthropic | 11.5% | source | — | — | |
| 26 | Claude-Opus-4 Supported Anthropic | 10.7% | source | — | — | |
| 27 | Gemini 3.1 Flash Lite Google | 8.6% | source | — | — | |
| 28 | GLM-4.5 Z.ai (Zhipu) | 8.3% | source | — | — | |
| 29 | GLM-4.5-Air Z.ai (Zhipu) | 8.1% | source | — | — | |
| 30 | o1-pro OpenAI | 8.1% | source | — | — | |
| 31 | Claude 3.7 Sonnet Supported Anthropic | 8.0% | source | — | — | |
| 32 | o1 Estimated OpenAI | 8.0% | source | — | — | |
| 33 | Claude Sonnet 4 Supported Anthropic | 7.8% | source | — | — | |
| 34 | Gemini 2.0 Flash Thinking (Jan 2025) Google | 6.6% | source | — | — | |
| 35 | Llama 4 Maverick Meta | 5.7% | source | — | — | |
| 36 | GPT-4.5 OpenAI | 5.4% | source | — | — | |
| 37 | GPT-4.1 OpenAI | 5.4% | source | — | — | |
| 38 | Gemini 1.5 Pro (Feb 2024) Google | 4.6% | source | — | — | |
| 39 | Mistral Medium 3 Mistral | 4.5% | source | — | — | |
| 40 | Amazon Nova Pro Amazon | 4.4% | source | — | — | |
| 41 | Claude-Sonnet-3.5 Estimated Anthropic | 4.1% | source | — | — | |
| 42 | Amazon Nova Lite Amazon | 3.6% | source | — | — | |
| 43 | GPT-4o (2024-08-06) Supported OpenAI | 2.7% | source | — | — |