From Perception to Action: leaderboard
Interactive 3D vision-reasoning benchmark where models plan physical actions in puzzle and stacking environments.
Metric: pass@1 (imported). Source: arxiv.org. Source: benchmarklist.com. Saturation forecast: Around March 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 22.9 | #110 |
| 2 | Gemini 3 Pro | 19.3 | #80 |
| 3 | Claude Opus 4.5 | 15.6 | #83 |
| 4 | Claude Sonnet 4.5 | 13.8 | #142 |
| 5 | Kimi K2.5 | 13.8 | #144 |
| 6 | Gemini 3 Flash | 11.9 | #99 |
| 7 | GPT-5 Mini | 11 | #186 |
| 8 | O3 | 10.1 | #124 |
| 9 | Seed-1.6 | 10.1 | #265 |
| 10 | Qwen 3 VL 8B (Thinking) | 8.3 | |
| 11 | GLM-4.6V | 7.3 | #316 |
| 12 | Seed 1.6 Flash | 7.3 | #436 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=from-perception-to-action · How It Works · Data refreshed daily, snapshot 2026-10-20.