The best AI image models, ranked
Every leading AI image generator and editor, ranked by blind side-by-side human votes on LMArena - overall and by category. 103 models, 14 leaderboards.
Text to Image
A written prompt in, an image out. 5,246,948 votes · 86 models
GPT-Image-2.5 Sunburst
OpenAI
1429±5
+18 pts ahead of #2 · 68,073 votes
GPT-Image-2.6 Sunburst (Medium)
OpenAI
1410±9
18 pts behind #1 · 6,547 votes
GPT-Image-2.5 Flare
OpenAI
1400±5
29 pts behind #1 · 67,372 votes
| Rank | Model | Score | Votes |
|---|---|---|---|
| 1 | GPT-Image-2.5 SunburstOpenAI · Proprietary | 1429 ±5 | 68,073 |
| 2 | GPT-Image-2.6 Sunburst (Medium)OpenAI · Proprietary | 1410 ±9 | 6,547 |
| 3 | GPT-Image-2.5 FlareOpenAI · Proprietary | 1400 ±5 | 67,372 |
| 4 | MAI-Image-2.7Microsoft AI · Proprietary | 1387 ±12 | 3,082 |
| 5 | GPT-Image-2 (Medium)OpenAI · Proprietary | 1382 ±4 | 110,696 |
| 6 | GPT-Image-2.6 Flare (Medium)OpenAI · Proprietary | 1375 ±9 | 6,447 |
| 7 | MAI-Image-2.6Microsoft AI · Proprietary | 1331 ±5 | 42,837 |
| 8 | Gemini Nano Banana 2.1Google · Proprietary | 1328 ±9 | 4,957 |
| 9 | Grok Imagine Image 2.0xAI · Proprietary | 1318 ±8 | 9,894 |
| 10 | Reve 2.1Reve · Proprietary | 1302 ±8 | 7,897 |
Leader by category
Image Edit
An image (or several) in, an edited image out. 27,128,416 votes · 64 models
GPT-Image-2.5 Sunburst
OpenAI
1519±4
+35 pts ahead of #2 · 140,743 votes
GPT-Image-2.5 Flare
OpenAI
1484±4
35 pts behind #1 · 137,820 votes
GPT-Image-2.6 Sunburst (Medium)
OpenAI
1481±7
38 pts behind #1 · 16,881 votes
| Rank | Model | Score | Votes |
|---|---|---|---|
| 1 | GPT-Image-2.5 SunburstOpenAI · Proprietary | 1519 ±4 | 140,743 |
| 2 | GPT-Image-2.5 FlareOpenAI · Proprietary | 1484 ±4 | 137,820 |
| 3 | GPT-Image-2.6 Sunburst (Medium)OpenAI · Proprietary | 1481 ±7 | 16,881 |
| 4 | MAI-Image-2.7Microsoft AI · Proprietary | 1476 ±11 | 3,007 |
| 5 | GPT-Image-2 (Medium)OpenAI · Proprietary | 1462 ±3 | 371,930 |
| 6 | GPT-Image-2.6 Flare (Medium)OpenAI · Proprietary | 1461 ±7 | 16,830 |
| 7 | Grok Imagine Image 2.0xAI · Proprietary | 1437 ±6 | 33,908 |
| 8 | MAI-Image-2.6Microsoft AI · Proprietary | 1427 ±5 | 82,671 |
| 9 | Gemini Nano Banana 2.1Google · Proprietary | 1427 ±6 | 12,041 |
| 10 | Grok Imagine Image 2.0 (20260801)xAI · Proprietary | 1419 ±4 | 144,482 |
Leader by category
About the AI Image Arena
What is the AI Image Arena leaderboard?
A ranking of AI image generators and image editors built from millions of blind, side-by-side human votes collected by LMArena (arena.ai). People compare two anonymous images made from the same prompt and pick the better one; the votes are turned into a score for every model. This page mirrors LMArena's published image leaderboards.
What is the best AI image generator right now?
The model ranked #1 on the Text to Image › Overall board. Rankings move every few weeks as new models launch, so check the date at the top of each board. For a specific job - photorealism, product and branding visuals, portraits or text in images - use that category's board, because leaders differ by task.
How are the scores calculated?
LMArena fits a Bradley-Terry model to the pairwise votes (the statistics behind Elo-style ratings) and reports each model's score with a 95% confidence interval. A higher score means people preferred that model's images more often. The rank spread shows every rank a model could plausibly hold given its interval - models whose intervals overlap are statistically tied.
What is the difference between Text to Image and Image Edit?
Text to Image boards start from a written prompt only. Image Edit boards start from one image (Single Image) or several images (Multi Image) plus an instruction - for example changing a background, swapping a garment or combining a product with a person.
Where does the data come from, and can I reuse it?
From LMArena's public leaderboard dataset on Hugging Face (lmarena-ai/leaderboard-dataset), published under CC BY 4.0, plus the two single-image edit category boards from arena.ai. Flashlook republishes it with attribution and does not change any score. You can reuse it under the same license - credit LMArena.
Which AI image model is best for fashion and product photos?
Look at the Product, Branding & Commercial Design and Photorealistic boards, and the Image Edit boards if you start from a product photo. For catalog work, raw image quality is only part of it: garments must stay exactly as made and the same model face has to hold across every SKU - that is the problem Flashlook is built for.
Data: LMArena (lmarena-ai/leaderboard-dataset), CC BY 4.0. Published Nov 17, 2026. Flashlook mirrors the rankings unchanged - about the data.

