Cybench — Epoch AI
Cybench — Epoch AI is run by Epoch AI. It has ranked 22 models, of which the catalogue holds 18, scoring from 0.05 to 0.93.
measured by Epoch AI · the board itself
| Place | Model | Metric | Score |
|---|---|---|---|
| 1stof 22 | Claude Opus 4.6 | Unguided % Solved | 0.93 |
| 2ndof 22 | Claude Opus 4.5 | Unguided % Solved | 0.82 |
| 3rdof 22 | Claude Sonnet 4.5 | Unguided % Solved | 0.6 |
| 6thof 22 | Claude Opus 4.1 | Unguided % Solved | 0.42 |
| 7thof 22 | Grok 4.1 | Unguided % Solved | 0.39 |
| 8thof 22 | Claude Opus 4 | Unguided % Solved | 0.38 |
| 9thof 22 | Claude Sonnet 4 | Unguided % Solved | 0.35 |
| 10thof 22 | Grok 4 Fast | Unguided % Solved | 0.3 |
| 11thof 22 | o3-mini | Unguided % Solved (medium) | 0.225 |
| 12thof 22 | Claude 3.7 Sonnet | Unguided % Solved | 0.2 |
| 14thof 22 | Claude 3.5 Sonnet | Unguided % Solved | 0.175 |
| 15thof 22 | GPT-4o (2024-11-20) | Unguided % Solved | 0.125 |
| 16thof 22 | Claude 3 Opus | Unguided % Solved | 0.1 |
| 17thof 22 | o1-mini | Unguided % Solved (medium) | 0.1 |
| 18thof 22 | o1 | Unguided % Solved | 0.1 |
| 19thof 22 | Mixtral-8x22B-Instruct-v0.1 | Unguided % Solved | 0.075 |
| 20thof 22 | Meta-Llama-3.1-405B-Instruct | Unguided % Solved | 0.075 |
| 22ndof 22 | Meta-Llama-3-70B-Instruct | Unguided % Solved | 0.05 |
Read from the board on 2026-09-16