Benchmarks › AA Image Arena
AA Image Arena
Generate or edit images for blind human preference comparisons
1,222top score · GPT Image 2.5 Sunburst
93models ranked
30open-weights models
Bradley-Terry Elohigher is better
Oct 20, 2026last updated
How the test works
1
Prompts
Human-curated rotation informed by anonymized user data
- What the model doesGenerate or edit images for blind human preference comparisons
- Kinds of tasksBlind pairwise preference comparison
- In → outText, reference image for editing → Image
2
Votes
Blind human pairwise preference votes
- How runs combineFit all eligible pairwise votes jointly with Bradley-Terry
3
Rating
Bradley-Terry Elo
- Better isHigher
- RefreshedMonthly prompt refresh
Sources: artificialanalysis.ai
AA Image Arena · Text-to-image leaderboardtop 10 of 93
Full leaderboard with filters →| Model | Arena Elo | Edits photos | Max output | Reference images | Price | Released |
|---|---|---|---|---|---|---|
1,222max | Yes | 1024x1024 | Yes | $0.21 / image | Sep 2026 | |
1,207max | Yes | 1024x1024 | Yes | $0.21 / image | Sep 2026 | |
1,186high | Yes | 1024x1024 | Yes | $0.21 / image | Apr 2026 | |
1,176unstated | Yes | 1K | Yes | $0.034 / image | Oct 2026 | |
1,166unstated | Yes | 1024x1024 | Yes | $0.039 / image | Aug 2026 | |
1,166unstated | Yes | 1024x1024 | Yes | $0.06 / image | Aug 2026 | |
1,140unstated | Yes | 1K | Yes | $0.067 / image | Feb 2026 | |
1,127unstated | Yes | 1024x1024 | Yes | $0.01 / image | Jul 2026 | |
1,123high | Yes | 1024x1024 | Yes | $0.13 / image | Dec 2025 | |
1,119 | Yes | 1024x1024 | Yes | $0.019 / image | Sep 2026 |
Versions and boards on Model Menu
| Board | Models |
|---|---|
AA Image Arena · Text-to-imageShown above | 93 |
| 69 |
AA Image Arena questions
What is AA Image Arena?›
Generate or edit images for blind human preference comparisons. Created by Artificial Analysis.
Which model is best on AA Image Arena?›
GPT Image 2.5 Sunburst leads AA Image Arena · Text-to-image with 1,222, ahead of GPT Image 2.5 Flare (1,207) and GPT Image 2 (1,186).
What is the best open-weights model on AA Image Arena?›
Qwen Image 2.1 with 1,042, then Ideogram 4.0 (1,026).
How is AA Image Arena scored?›
Bradley-Terry Elo; blind human pairwise preference votes; 1 run per task per task.
Which versions of AA Image Arena are there?›
AA Image Arena · Text-to-image (93 models), AA Image Arena · Editing (69 models).