SudokuBench (Multi-step, 4x4): leaderboard
Tool-free Sudoku variant reasoning: percentage of complete correct puzzle solutions. Interactive cell placements terminate on an incorrect move or other termination condition. Source-specific grid cohort; four-decimal published solve fractions. Explicit Preview and Beta API checkpoint identities retained.
Metric: 4x4 Solve Rate (%). Source: pub.sakana.ai. Saturation forecast: Estimated already saturated. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 (High) | 80 | #98 (GPT-5) |
| 2 | Gemini 2.5 Pro (Preview 03-25) | 73.33 | #123 |
| 3 | GPT-5 (Medium) | 66.7 | #98 (GPT-5) |
| 4 | DeepSeek R1 | 60 | #251 |
| 5 | O3 Mini (High) | 60 | #272 (O3 Mini) |
| 6 | Grok 3 Mini Beta | 53.33 | #296 |
| 7 | GPT-5 (Low) | 53.3 | #98 (GPT-5) |
| 8 | Qwen 3 235B A22B | 40 | #313 |
| 9 | Claude 3.7 Sonnet (Thinking) | 40 | #249 (Claude 3.7 Sonnet) |
| 10 | QwQ-32B | 26.67 | #421 |
| 11 | GPT-4.1 | 13.33 | #248 |
| 12 | Qwen 3 30B A3B | 6.67 | #493 |
| 13 | Llama 4 Maverick | 0 | #464 |
| 14 | Llama 4 Scout | 0 | #656 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sudokubench-multi-step-4x4 · How It Works · Data refreshed daily, snapshot 2026-10-20.