SceneBench: leaderboard
Metric: Average accuracy (%) over SceneBench's six tasks (title and comment prediction, clip QA, scene QA with and without audio cues, inverse VQA) on SceneBench (2,485 long videos averaging 33 minutes), four-option multiple choice, each model at its own recommended frame budget; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 80.3 | #145 |
| 2 | Kimi K2.5 | 73 | #139 |
| 3 | Gemini 2.5 Flash Lite | 53.8 | #413 |
| 4 | Qwen 2.5 VL 7B Instruct | 50.5 | #643 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=scenebench · How It Works · Data refreshed daily, snapshot 2026-10-11.