RE-Edit - Causal: leaderboard
RE-Edit (2026) tests instruction-based image editors on 1,000 curated requests whose correct edit depends on implicit reasoning. This board is the pass rate on the causal-reasoning subset (the causes and effects an edit implies), with a Qwen3-VL-30B judge marking each edited image pass or fail against the case rationale. Higher is better.
Metric: Pass rate (%) on the causal-reasoning subset of RE-Edit's 1,000 curated edit requests (cause and effect of the edit): a Qwen3-VL-30B judge marks each edited image pass or fail against the case rationale; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Nano Banana | 15.5 | |
| 2 | FLUX.2-dev | 15 | |
| 3 | Seedream 4.0 | 14.5 | |
| 4 | Qwen-Image-Edit | 13 | |
| 5 | DreamOmni2 | 9 | |
| 6 | FLUX.1 Kontext [dev] | 7.5 | |
| 7 | Step1X-Edit-v1p1 | 7.5 | |
| 8 | OmniGen2 | 5.5 | #4321 |
| 9 | HiDream-E1 | 5.5 | |
| 10 | Step1X-Edit-v1p2-preview | 5 |
Interactive version: theaggregate.ai/benchmark?slug=re-edit-causal · How It Works · Data refreshed daily, snapshot 2026-11-13.