EgoSafetyBench: leaderboard
Metric: Video-level balanced accuracy (%; mean of the recall on unsafe videos and on safe videos over the 800 situational-track scenarios, a video called unsafe when any of its chunks is flagged, macro-averaged over videos; ten VLM safety guards judging egocentric robot videos split into ten half-second chunks (ten frames each at 1280x720), each chunk judged alone under the current-chunk protocol with a single-word verdict and greedy decoding; higher is better). Source: arxiv.org. Saturation forecast: Around December 2026. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Flash Lite | 87.3 |
| 2 | Qwen 3.5 4B | 85.5 |
| 3 | Gemini 3.5 Flash | 85.4 |
| 4 | Qwen 3 VL 4B Instruct | 84.9 |
| 5 | InternVL3.5-8B | 82.7 |
| 6 | Claude Sonnet 4.6 | 82.5 |
| 7 | Gemma 3 4B (IT) | 65.3 |
| 8 | Qwen 3.5 0.8B | 65.3 |
Interactive version: theaggregate.ai/benchmark?slug=egosafetybench · How It Works · Data refreshed daily, snapshot 2026-09-29.