VideoGAIA: leaderboard
VideoGAIA (2026) is an agentic video-understanding benchmark of 271 human-verified tasks in which a model inspects a video, searches the web, visits pages and re-samples frames over up to 40 steps to answer a question. Pass@1 accuracy in percent against the reference answer, judged by GPT-5.5 and averaged over three runs; higher is better. All models share one ReAct agent loop, tools and frame budget.
Metric: Pass@1 accuracy (%; all 271 tasks; unified ReAct agent loop: 20 uniformly sampled frames plus web search, page visit and a thinking-with-videos tool that samples up to 20 more frames from a chosen segment, at most 40 steps; answers judged against the human-verified reference by GPT-5.5 (DeepSeek-V4-Pro fallback); mean of three runs). Source: arxiv.org. Saturation forecast: Not forecast. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Seed 2.0 Pro | 58.3 | #209 |
| 2 | Qwen 3.7 Plus | 56.09 | #203 |
| 3 | Kimi K3 | 54.61 | #53 |
| 4 | GPT-5.5 | 53.14 | #48 |
| 5 | Kimi K2.6 | 52.77 | #173 |
| 6 | GPT-5.4 | 52.03 | #121 |
| 7 | Qwen 3.5 Plus | 52.03 | #227 |
| 8 | Gemini 3.1 Pro (Preview) | 50.92 | #91 |
| 9 | GPT-5.2 | 50.92 | #172 |
| 10 | Qwen 3.5 397B A17B | 50.18 | #242 |
| 11 | Qwen 3.8 Max | 50.18 | #55 |
| 12 | GLM-5V Turbo | 50.18 | #342 |
| 13 | Kimi K2.5 | 49.82 | #252 |
| 14 | Qwen 3.6 Plus | 49.08 | #217 |
| 15 | MiMo-V2.5 | 48.71 | #353 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=videogaia · How It Works · Data refreshed daily, snapshot 2026-11-12.