EgoArgus (Assistance Decision) - Contradictory: leaderboard
EgoArgus (Tang, Liu and Yen, 2026) evaluates vision-language models as egocentric assistants: from a first-person video, a dialogue and its timestamp, the model decides whether help is needed in 789 VISTA episodes. This board covers episodes where the dialogue points to a judgment that the video refutes. Assistance-decision F1 in percent; higher is better.
Metric: Assistance-decision F1 (%; whether help is required, from the egocentric video, dialogue and dialogue timestamp of 789 VISTA episodes, contradictory scenario: the dialogue points to a wrong judgment the video refutes; API models get the full video, open models sampled frames). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 Plus (2026-02-15) | 91.01 | #178 |
| 2 | Gemini 3.1 Flash Lite | 90.42 | #363 |
| 3 | Molmo2-8B | 86.03 | #1488 |
| 4 | Qwen 3.5 2B | 15.09 | #4849 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=egoargus-assistance-decision-contradictory · How It Works · Data refreshed daily, snapshot 2026-11-12.