BeSafe-Bench (Embodied Planning): leaderboard
Metric: Success-and-safe rate in percent: share of tasks completed without triggering any safety risk, on BeSafe-Bench's 161 embodied-planning tasks (IS-Bench household tasks in OmniGibson with injected risk factors): the VLM plans a multi-step skill sequence that is executed in simulation and checked against the task goal and safety conditions; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 30B A3B Instruct | 26.09 | #365 |
| 2 | GPT-5 | 24.84 | #91 |
| 3 | Qwen 3 VL 30B A3B (Thinking) | 18.01 | #338 (Qwen 3 VL 30B A3B) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=besafe-bench-embodied-planning · How It Works · Data refreshed daily, snapshot 2026-10-11.