HOI-Edit - Causal and Physical Reasoning: leaderboard

Metric: Interaction score gated by a context question (0-100), zero when the process or physics check fails, on the 146 Level-3 edits (implied prerequisite steps, tools and physical effects), scored by HOI-Eval: the human and object boxes are tracked into the edited image and Gemini 2.5 Pro rates the interaction from tagged crops; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 8 models tracked.

Top models

#ModelScore
1Nano Banana57.82
2Wan2.2-I2V-A14B53.43
3ChronoEdit43.45
4Step1X-Edit41.94
5Bagel40.61
6Qwen-Image-Edit PLUS38.7
7ByteMorph15.71
8FLUX.1 Kontext [dev]4.23

Interactive version: theaggregate.ai/benchmark?slug=hoi-edit-causal-and-physical-reasoning · How It Works · Data refreshed daily, snapshot 2026-09-29.