EG-VQA: leaderboard
EG-VQA (2026) tests evidence-grounded video question answering: models answer open-ended descriptive, temporal, causal and counterfactual questions about videos and must also point to the supporting moments. The score is strict answer accuracy judged by an LLM, in percent; higher is better.
Metric: Strict accuracy (%): share of answers the judge rates fully correct, all four question types, on the 2,889-question EG-VQA test split (open-ended questions over ActivityNet Captions, HiREST and YouCook2 videos, 64 uniformly sampled frames), answers graded by a Gemini-2.5-Pro judge against the reference answer; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 28.47 | #741 |
| 2 | Gemini 2.5 Flash | 24.87 | #453 |
| 3 | Qwen 2.5 VL 7B Instruct | 14.31 | #1833 |
| 4 | InternVL3-8B | 11.71 | #1688 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=eg-vqa · How It Works · Data refreshed daily, snapshot 2026-11-13.