OpenRouter · Vercel · CC BY 4.0
Arena · CC BY 4.0
People compare two images made from the same prompt, blind, and vote. It ranks how well a model creates, not how much it’s used.
| Rank | Model | Quality score | Votes |
|---|---|---|---|
| 1 | GPT Image 2.5 Sunburst OpenAI | 1424 pts | 10,884 |
| 2 | GPT Image 2.5 Flare OpenAI | 1401 pts | 10,058 |
| 3 | GPT Image 2 OpenAI | 1383 pts | 88,744 |
| 4 | MAI Image 2.6 Microsoft AI | 1335 pts | 18,625 |
| 5 | Reve 2.1 Reve | 1302 pts | 8,305 |
| 6 | Grok Imagine Image 2.0 SpaceXAI | 1301 pts | 7,257 |
| 7 | Muse Image Meta | 1276 pts | 35,422 |
| 8 | Reve 2.0 Reve | 1269 pts | 15,802 |
| 9 | Nano Banana 2 Google | 1261 pts | 53,228 |
| 10 | Seedream 5.0 Pro ByteDance | 1256 pts | 85,690 |
| 11 | Qwen Image 3.0 Pro Alibaba | 1256 pts | 11,520 |
| 12 | MAI Image 2.5 Microsoft AI | 1254 pts | 61,448 |
Source: Arena image leaderboard (lmarena-ai/leaderboard-dataset), released Sep 24, 2026 · CC BY 4.0 · Rank changes (▲▼) compare with the previous release