VisReason (Vision-Centric) - Sudoku Solving: leaderboard

VisReason Sudoku Solving is a structural-reasoning category of VisReason (2026): the model must solve visually presented Sudoku variants. The score is accuracy in percent, higher is better.

Metric: Accuracy (%) on the Sudoku Solving category (structural reasoning: solve visually presented Sudoku variants) of VisReason, answers scored by format (regular-expression match, GPT-5-mini judge for open-ended answers, IoU above 0.5 for boxes), zero-shot with chain-of-thought prompt templates, one per answer format; rows not shaded gray in the paper use explicit reasoning (GPT-5 and Gemini 3 Pro at low reasoning effort); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 20 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-5.2Thinking38.2#185
2Gemini 3 Pro (Preview)Low30#119
3GPT-5Low22.5#178
4O4 Mini15.5#368
5GPT-5 MiniLow15#364
6Qwen 3 VL 8B Instruct9.2#1029
7GPT-5 Nano9#906
8InternVL3-8B8.5#1782
9Qwen 3 VL 30B A3B (Thinking)8.5#719
10Qwen 2.5 VL 32B Instruct8#1062
11GPT-4o7.5#751
12Qwen 3 VL 32B (Thinking)7#573
13Gemini 2.5 Flash6#463
14Qwen 3 VL 8B (Thinking)6#910
15Keye-VL-1.5-8BThinking4.5#1720

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=visreason-vision-centric-sudoku-solving · How It Works · Data refreshed daily, snapshot 2026-11-21.