EgoMemReason - Cumulative State Tracking: leaderboard
EgoMemReason task (Wang et al., COLM 2026) in which the model must say how an object's location or condition changed across observations hours or days apart in week-long egocentric video, using only footage before the question time. Accuracy in percent on 100 multiple-choice questions; higher is better.
Metric: Accuracy (%; 100 multiple-choice questions). Source: arxiv.org. Saturation forecast: Around 2029. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 46 | #159 |
| 2 | Gemini 3.1 Pro (Preview) | 40 | #91 |
| 3 | Molmo2-8B | 36 | #1488 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=egomemreason-cumulative-state-tracking · How It Works · Data refreshed daily, snapshot 2026-11-12.