ChronoPhyBench - Frame Sorting (4 Frames): leaderboard
ChronoPhyBench (2026) tests whether multimodal models predict what happens next in real physical-interaction videos rather than relying on language priors. On this board the model sees eight frames of a video with its caption and puts four held-out frames in the order the physical process produces them; the score is the percent of exactly correct orders (chance 4.2 percent). Higher is better.
Metric: Accuracy (%) on ChronoPhyBench chronological sorting of 4 held-out frames: given the video context and caption, put the frames in the order the physical process produces them (exact order required); chance 4.2; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Doubao-Seed-1-8-251228 | 46.5 | #282 |
| 2 | Doubao-Seed-2.0-Pro-260215 | 45.5 | #152 |
| 3 | GPT-5.5 | 40.5 | #49 |
| 4 | Gemini 3 Pro (Preview) | 35.5 | #113 |
| 5 | GPT-5.3 | 33.5 | #197 |
| 6 | Doubao-Seed-1.6-thinking-250715 | 27 | #324 |
| 7 | Qwen 3 VL 32B Instruct | 23 | #585 |
| 8 | GPT-5.4 | 16 | #123 |
| 9 | Qwen 3 VL 8B (Thinking) | 16 | #887 |
| 10 | Qwen 3 VL 8B Instruct | 14 | #981 |
| 11 | GLM-4.1V-9B (Thinking) | 13.5 | #1060 (GLM-4.1V-9B) |
| 12 | InternVL3-38B | 8 | #903 |
| 13 | InternVL3-14B | 8 | #1182 |
| 14 | InternVL3-8B | 5 | #1688 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=chronophybench-frame-sorting-4-frames · How It Works · Data refreshed daily, snapshot 2026-11-13.