MazeBench ASCII with Python Rooms Visited
Games · 2026-07-28
Exploration score from the same Python-enabled MazeBench ASCII runs: the share of the world's 255 non-starting rooms the agent visited, as the leaderboard displays it. It measures exploration within the same run rather than puzzle mastery, so it is kept as a diagnostic.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Astra | 43.9 |
| GPT-6.1 Sol | 39.6 |
| Claude Opus 5.5 | 26.3 |
| Claude Sonnet 5.5 | 18.8 |
| GPT-6 Sol | 14.5 |