Spatial Competence Benchmark - Planning: leaderboard

The 39 planning subtasks of the Spatial Competence Benchmark (SCBench, 2026), such as 3D mazes, fluid simulation and terrain levelling, with plans executed in simulators, single-turn and without tools. Accuracy in percent; higher is better.

Metric: Accuracy (%) over the 39 planning subtasks (3D maze, hyper-snake, fluid simulation, terrain levelling) of the Spatial Competence Benchmark (SCBench) subtasks with executable outputs checked by deterministic verifiers or simulators, each subtask scored in [0, 1] (binary or partial credit), every subtask weighted equally; single-turn, zero-shot, one sample, no tools, no self-correction; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 3 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-5.2xHigh50#185
2Gemini 3 Pro (Preview)39#119
3Claude Sonnet 4.5Thinking27.5#261

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

Interactive version: theaggregate.ai/benchmark?slug=spatial-competence-benchmark-planning · How It Works · Data refreshed daily, snapshot 2026-11-21.