Reasoning on Sudoku Extreme
99.8Pass@1 AccuracyEqR (+ inference scaling)
Evaluation results
| Method | Links | |
|---|---|---|
| EqR (+ inference scaling)Backbone=MLP-mixer, recipe=inference scaling2026.05 | 99.8 | |
| EqR (+ inference scaling)Backbone=Transformer, recipe=inference scaling2026.05 | 95.9 | |
| Attractor Model# Parameters=27M2026.05 | 91.4 | |
| EqR (+ training intervention)Backbone=MLP-mixer, recipe=training intervention2026.05 | 86.4 | |
| BaselineBackbone=MLP-mixer2026.05 | 84.1 | |
| TRM# Parameters=7M2026.05 | 74.7 | |
| EqR (+ training intervention)Backbone=Transformer, recipe=training intervention2026.05 | 74.7 | |
| BaselineBackbone=Transformer2026.05 | 72 | |
| SR²Params=3.4M2025.10 | 66.63 | |
| HRMParams=27.3M2025.10 | 55 | |
| HRM# Parameters=27M2026.05 | 55 | |
| Attractor Model# Parameters=7M2026.05 | 54.3 | |
| Reflective ModelParams=27.3M2025.10 | 53.12 | |
| Recurrent DepthParams=3.4M2025.10 | 42.52 | |
| TransformerParams=27.3M2025.10 | 1.17 | |
| Block Universal TransformerParams=3.4M2025.10 | 0 | |
| Deepseek R1# Parameters=671B2026.05 | 0 | |
| Claude 3.7# Parameters=?2026.05 | 0 | |
| O3-mini-high# Parameters=?2026.05 | 0 | |
| Transformer# Parameters=27M2026.05 | 0 | |
| TRM# Parameters=27M2026.05 | 0 |