Median tokens (solved only)
Solve rate
| Model | n | Solved | Solve % | Median | Mean | Note |
|---|---|---|---|---|---|---|
| Kimi-K3 | 10 | 9 | 90% | 10 | 12 | best full-grid: solve + speed |
| GLM-5.2 | 10 | 9 | 90% | 31 | 34.1 | solves, but slowest median |
| MiniMax-M3 | 8 | 7 | 87.5% | 21 | 23.6 | 2 missing cells |
| Gemma-4-31B | 10 | 7 | 70% | 13 | 10.9 | fast when it hits (+20pp vs v2) |
| gpt-oss-120b | 10 | 7 | 70% | 14 | 15 | full grid now; morphology kills 3 |
| Mistral-Med | 10 | 4 | 40% | 15 | 15.2 | doubled solve rate vs v2 |
| MiMo-V2.5 | 10 | 4 | 40% | 28.5 | 31 | synonym + morph collapses |