VideoNet - Binary Zero-Shot: leaderboard
Binary setting of VideoNet (Yadav et al., CVPR 2026): given an action name and its domain, the model reasons and then answers yes or no to whether a clip shows that action, over two verified positive and two hard-negative clips per action and with no example clips. Accuracy in percent (chance is 50); higher is better.
Metric: Binary Accuracy (%; does the clip show the named action, 2 positive and 2 hard-negative clips per action, no examples). Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 (Medium) | 72.9 | #163 (GPT-5) |
| 2 | GPT-5.4 (2026-03-05) (Medium) | 72.4 | #84 (GPT-5.4 (2026-03-05)) |
| 3 | Gemini 3.1 Pro (Preview) (High) | 72 | #91 (Gemini 3.1 Pro (Preview)) |
| 4 | Gemini 3 Flash (Preview) (High) | 70.3 | #141 (Gemini 3 Flash (Preview)) |
| 5 | Qwen 3 VL 8B Instruct | 59.3 | #982 |
| 6 | Molmo2-8B | 55.8 | #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-binary-zero-shot · How It Works · Data refreshed daily, snapshot 2026-11-12.