PhyEditBench - Physical Plausibility: leaderboard
PhyEditBench (2026) tests physics-aware instruction-based image editing on real-world physical processes extracted from videos, with outputs scored by a GPT-4o judge from 1 to 10 on consistency, instruction following, physical plausibility and image quality. The board is the physical plausibility score alone on the 238 real-world instances, averaged over the five run types; higher is better.
Metric: Physical plausibility judge score (1-10 scale) on the 238 real-world physical-process instances, averaged over the five run types; GPT-4o judge scores each output 1-10 on consistency, instruction following, physical plausibility and image quality against the instruction, physical explanation, invariants and reference frames; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | ChronoEdit-14B | 8.42 | |
| 2 | GPT-Image-1.5 | 8.3 | |
| 3 | Seedream4.0 | 8.22 | |
| 4 | UniWorld-V2 | 6.73 | |
| 5 | gemini-2.5-flash-image | 6.64 | |
| 6 | Qwen-Image-Edit | 6.37 | |
| 7 | Step1X-Edit-v1p1 | 6.19 | |
| 8 | BAGEL-7B-MoT (Thinking) | 5.96 | #5470 (BAGEL-7B-MoT) |
| 9 | BAGEL-7B-MoT | 5.69 | #5470 |
| 10 | InstructPix2Pix | 5.58 |
Interactive version: theaggregate.ai/benchmark?slug=phyeditbench-physical-plausibility · How It Works · Data refreshed daily, snapshot 2026-11-13.