EgoSAT - Short-Horizon Recall: leaderboard
Metric: Multiple-choice accuracy (%) on the action a short time ago; EgoSAT: streaming questions on Ego4D egocentric video answered online from the clips seen so far (strict prefix truncation at each query time); four options per question (the answer, two temporally adjacent hard negatives and one absurd negative), chance 25; higher is better. Source: arxiv.org. Saturation forecast: Around 2032. 9 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 2.5 VL 72B Instruct | 51.87 |
| 2 | Qwen 2.5 VL 32B Instruct | 50.86 |
| 3 | Qwen 2.5 VL 7B Instruct | 42.79 |
| 4 | Claude Sonnet 4 | 40.46 |
| 5 | Gemini 2.5 Pro | 39.8 |
Interactive version: theaggregate.ai/benchmark?slug=egosat-short-horizon-recall · How It Works · Data refreshed daily, snapshot 2026-09-29.