CapRiCorn-1K-V: leaderboard
The visual-only track of CapRiCorn-1K (Chen et al., 2026): captioners that see frames without audio describe the 1,000 videos and are scored on the vision keypoints. Keypoint accuracy is the share of annotated visual keypoints a GPT-4.1 judge finds mentioned, in percent; higher is better.
Metric: Keypoint accuracy (%; share of vision-only keypoints a GPT-4.1 judge marks correctly mentioned in the caption, frames only, 1,000 videos). Source: arxiv.org. Saturation forecast: Around August 2028. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 27B | 13.4 | #339 |
| 2 | Qwen 3.5 122B A10B | 11.7 | #329 |
| 3 | Qwen 3.6 35B A3B | 11.7 | #461 |
| 4 | Qwen 3.5 9B | 10.7 | #803 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=capricorn-1k-v · How It Works · Data refreshed daily, snapshot 2026-11-12.