Blind: prompts used no knowledge of the challenge code or other players' solutions.Honor system. Every prompt can be reviewed, so anyone can check.
Inference time depends on the hardware used.
| Rank | Model | Inference | Quant | Prompts | Player | Hardware | Run |
|---|---|---|---|---|---|---|---|
| 1 | devstral-small-2:24b | 45.0 s32,146 tokens | Q4_K_M | 1 | ShapeThistle4673 | Apple M4 Max, 48 GB | View |
| 2 | gpt-oss:20b thinking | 49.9 s120,835 tokens | MXFP4 | 1 | TiltCedar6543 | Apple M4 Max, 48 GB | View |
| 3 | qwen3.8:27b thinking | 1 m 40 s19,007 tokens | Q4_K_M | 1 | KeepQuilt1658 | Apple M4 Max, 48 GB | View |
| 4 | qwen3.6:27b thinking | 3 m 42 s63,595 tokens | Q4_K_M | 1 | HikePearl133 | Apple M4 Max, 48 GB | View |
| 5 | north-mini-code-1.0 thinking | 3 m 57 s112,102 tokens | Q4_K_M | 1 | RestJade7528 | Apple M4 Max, 48 GB | View |
| 6 | qwen3.6:35b thinking | 8 m 31 s184,139 tokens | Q4_K_M | 5 | PluckMoon3277 | Apple M4 Max, 48 GB | View |