EgoPro-Bench - Action: leaderboard
Proactive alerting in egocentric video streams, from EgoPro-Bench (Ran et al., 2026): processing the stream frame by frame, the model must raise an alert while a target scene event or human action is happening and stay silent otherwise. Frame-level F1 in percent over 200 test videos; higher is better.
Metric: F1 (%; alert timing for scene events and human actions, 200 test videos; streaming frame-by-frame protocol with history, the model answers Attention or Silence at every frame sampled at 1 fps). Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 4B Instruct | 52.24 | #1281 |
| 2 | Qwen 3 VL 30B A3B Instruct | 49.47 | #870 |
| 3 | Qwen 3 VL 8B Instruct | 34.6 | #982 |
| 4 | Qwen 2.5 VL 72B Instruct | 33.72 | #845 |
| 5 | Qwen 2.5 VL 32B Instruct | 14.67 | #1055 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=egopro-bench-action · How It Works · Data refreshed daily, snapshot 2026-11-12.