Highlights
10 imagesBest models, by
Best Image Editing Models by Price
1 model without pricing omitted.
Best Image Editing Models by Speed
2 models waiting for enough speed data.
Best AI Models for Prompt Adherence
| # | Model | Elo |
|---|---|---|
| 1 | 1268 | |
| 2 | 1245 | |
| 3 | 1241 | |
| 4 | 1226 | |
| 5 | 1219 | |
| 6 | 1215 | |
| 7 | 1214 | |
| 8 | 1205 | |
| 9 | 1204 | |
| 10 | 1200 |
| 11 | 1198 | |
| 12 | 1198 | |
| 13 | 1197 | |
| 14 | 1190 | |
| 15 | 1188 | |
| 16 | 1188 | |
| 17 | 1145 | |
| 18 | 1144 | |
| 19 | 1140 | |
| 20 | 1115 | |
| 21 | 1084 |
GPT Image 1.5 leads the category with a 1268 Elo and 70.0% win rate, though the more expensive GPT Image 2 maintains a higher individual win rate of 85.7%. For cost-conscious users, FLUX.2 [klein] 9B offers a competitive 1219 Elo at a significantly lower price tier and faster generation speed than the top-ranked Google and OpenAI models.
Highlighted challenges
Man and Car in California
Combines two references (a person and a car) into one believable driving scene. It checks whether the model keeps the man’s identity and the car’s appearance while placing them convincingly on the coastline with matched lighting and perspective.
Bald man challenge
A targeted edit test. Adding a realistic head of hair requires a believable hairline and texture while keeping the face, expression, and lighting unchanged. Common failures are wig-like or pasted-on hair and subtle changes to the underlying face.
Keep the arena honest
Cast your vote
Pick winners in blind matchups. Every vote nudges the Elo and shapes these rankings.
Cast Your VoteSuggest a prompt
Got an idea worth testing? Submit a prompt and watch the models battle it out.