LMGame-Bench Sokoban — leaderboard

LLM game-playing benchmark: Sokoban puzzle testing multi-step planning and spatial reasoning.

Metric: Score. Source: huggingface.co. Status: saturation imminent. 25 models tracked.

Top models

#ModelScore
1GPT-5 (High)11
2O3 (2025-04-16)8
3Grok 46.3
4Grok 3 Mini Beta5.7
5O4 Mini (2025-04-16)5.3
6DeepSeek R1 05284.7
7Gemini 2.5 Pro (Preview 05-06)4.3
8GPT-OSS-120B4
9GPT-OSS-20B4
10Claude Opus 4 (20250514)4
11Claude Sonnet 4 (20250514)3
12Claude 3.7 Sonnet (20250219)2.3
13O1 (2024-12-17)2.3
14Qwen 3 235B A22B FP82.3
15Gemini 2.5 Flash (Preview 04-17)1.7

Interactive version: theaggregate.ai/benchmark?slug=lmgame-bench-sokoban · How the rankings work · Data refreshed daily, snapshot 2026-07-22.