SPIN-Bench - Planning: leaderboard
SPIN-Bench classical planning accuracy (Plan Acc): the percentage of PDDL planning problems a model solves, drawn from 21 domains whose tasks span fact retrieval, spatial reasoning and multi-step plans over growing state spaces. Higher is better, 0-100 (%).
Metric: PDDL plan accuracy (%). Source: spinbench.github.io. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O1 | 58.59 | #190 |
| 2 | O3 Mini | 51.25 | #282 |
| 3 | O4 Mini | 46.79 | #185 |
| 4 | DeepSeek R1 | 44.3 | #262 |
| 5 | Claude 3.5 Sonnet | 20.55 | #362 |
| 6 | O1 Mini | 13.2 | #334 |
| 7 | Llama 4 Maverick | 13.05 | #486 |
| 8 | GPT-4o | 8.75 | #357 |
| 9 | Llama 3.3 70B | 5.78 | #578 |
| 10 | GPT-4 Turbo | 5.62 | #417 |
| 11 | Claude 3.5 Haiku | 4.22 | #571 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=spin-bench-planning · How It Works · Data refreshed daily, snapshot 2026-11-05.