EgoSafetyBench - Visual-Channel Mismatch Detection: leaderboard
Metric: F1 (%) of flagging chunks whose in-scene sign, sticker or label misrepresents the physical situation, pooled over the chunks of the 400-scenario visual-channel track of paired misleading and truthful signs; 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 | Claude Sonnet 4.6 | 85 |
| 2 | Gemini 3.5 Flash | 82.1 |
| 3 | Gemini 3.1 Flash Lite | 80.5 |
| 4 | Qwen 3.5 4B | 67.8 |
| 5 | Qwen 3 VL 4B Instruct | 66.3 |
| 6 | Gemma 3 4B (IT) | 53 |
| 7 | Qwen 3.5 0.8B | 39 |
| 8 | InternVL3.5-8B | 36.8 |
Interactive version: theaggregate.ai/benchmark?slug=egosafetybench-visual-channel-mismatch-detection · How It Works · Data refreshed daily, snapshot 2026-09-29.