EG-VQA - Descriptive: leaderboard

Metric: Strict accuracy (%) on descriptive questions (directly observable actions, objects and states), 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: Around June 2028. 9 models tracked.

Top models

#ModelScore
1GPT-4o18.9
2Gemini 2.5 Flash14.58
3Qwen 2.5 VL 7B Instruct9.91
4InternVL3-8B9.71

Interactive version: theaggregate.ai/benchmark?slug=eg-vqa-descriptive · How It Works · Data refreshed daily, snapshot 2026-09-29.