Benchmarks / Agentic
MazeBench
Uses gems collected on the populated first-party ASCII-with-tools leaderboard. Score: share of MazeBench's 100 gems collected. Collected by Jake Boggs' benchmark dashboard.
MazeBench
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | GPT-6 Astra | max | 23.0% |
| 2 | GPT-6.1 Sol | max | 17.0% |
| 3 | Opus 5.5 | max | 13.0% |
| 4 | GPT-5.6 Sol | max | 13.0% |
| 5 | Opus 5 | max | 12.0% |
| 6 | Fable 5 | max | 11.0% |
| 7 | Fable 5.1 | max | 10.0% |
| 8 | Sonnet 5.5 | max | 8.0% |
| 9 | GPT-6 Sol | max | 7.0% |
| 10 | GPT-5.6 Terra | max | 6.0% |
| 11 | Gemini 3.8 Flash | max | 4.0% |
| 12 | Haiku 5.5 | max | 1.0% |
| 13 | Gemini 3.1 Pro | high | 1.0% |
| 14 | Gemini 3.7 Flash | max | 1.0% |
| 15 | Grok 4.6 | max | 1.0% |
| 16 | Kimi K3 | max | 1.0% |
| 17 | Sonnet 5 | max | 0.0% |
| 18 | DeepSeek V4 Flash 0731 | max | 0.0% |
| 19 | DeepSeek V4 Pro | max | 0.0% |
| 20 | Gemini 3.6 Flash | xhigh | 0.0% |
| 21 | GLM 5.3 | max | 0.0% |
| 22 | GLM 5.3 Flash | max | 0.0% |
| 23 | GPT-5.6 Luna | max | 0.0% |
| 24 | Muse Spark 1.3 | max | 0.0% |
| 25 | Qwen 3.8 Max | max | 0.0% |