VideoNet: leaderboard
Domain-specific action recognition in video (VideoNet, Yadav et al., CVPR 2026): the model sees a short trimmed clip from one of 37 domains such as figure skating, suturing or crochet and picks which of four same-domain actions it shows, where the wrong options are hard negatives that differ in subtle motion. Accuracy in percent on 4,000 test clips; higher is better.
Metric: Multiple-Choice Accuracy (%; one of four same-domain actions, hard negatives, 4,000 test clips). Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (High) | 69.9 | #91 (Gemini 3.1 Pro (Preview)) |
| 2 | Gemini 3 Flash (Preview) (High) | 68.7 | #141 (Gemini 3 Flash (Preview)) |
| 3 | GPT-5.4 (2026-03-05) (Medium) | 68 | #84 (GPT-5.4 (2026-03-05)) |
| 4 | GPT-5 (Medium) | 67.6 | #163 (GPT-5) |
| 5 | Qwen 3 VL 8B Instruct | 45 | #982 |
| 6 | Molmo2-8B | 44.9 | #1488 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=videonet · How It Works · Data refreshed daily, snapshot 2026-11-12.