HAVEN - Keyframe Selection: leaderboard
Keyframe selection on HAVEN (2026), a video understanding benchmark with aligned frames, shots, descriptions and full video documents: the model summarizes a video by choosing key frames within a 15 percent budget, scored by F1 against human-annotated keyframes in percent; higher is better.
Metric: Keyframe selection F1 (%) against human-annotated keyframes when summarizing a video into frames (video-to-video), 15 percent frame budget, zero-shot on a random sample of 100 videos per source dataset; higher is better. Source: arxiv.org. Saturation forecast: Around May 2027. 4 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | VideoLLaMA3-7B | 29.3 | #2064 |
| 2 | InternVideo2.5-Chat-8B | 27.3 | #2049 |
| 3 | Qwen 2.5 VL 7B Instruct | 20.9 | #1919 |
| 4 | Qwen 3 VL 8B Instruct | 17.3 | #1043 |
No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=haven-keyframe-selection · How It Works · Data refreshed daily, snapshot 2026-11-22.