Image Edit Arena Leaderboard
The latest AI image editing model leaderboard based on Image Edit Arena anonymous user voting. Covers Elo scores, confidence intervals, and vote counts for GPT-Image-1, Gemini, FLUX, Qwen, and more.
Top Model
GPT-image-2 (medium)
Top Score
1,465
Model Count
49
Data version
2026年06月03日
Data source: LM Arena
About This Leaderboard
This leaderboard ranks AI image editing models by quality. Data comes from LMArena's Image Edit Arena track, where users evaluate anonymous model outputs through blind voting.
Methodology Overview
Blind testing: Users submit editing instructions, two anonymous models generate results, and users vote for the better output.
Elo scoring: Uses the same Bradley-Terry model as text generation to scientifically reflect each model's relative strength.
Diverse editing scenarios: Covers style transfer, local editing, background replacement, object addition/removal, and more.
DataLearner provides in-depth analysis on top of the raw data, linking leaderboard models to the DataLearner model database so you can quickly access model details, API pricing, benchmark scores, and more.
Ranking Table
| Rank (UB) | Model | Score | 95% CI | Votes | Organization | License |
|---|---|---|---|---|---|---|
 GPT-image-2 (medium)OpenAI | 1,465 | +/-5 | 79,498 | OpenAI | Proprietary | |
 mai-image-2.5Microsoft AI | 1,401 | +/-8 | 6,493 | Microsoft AI | Proprietary | |
| GPT Image 1OpenAI | 1,390 | +/-4 | 393,343 | OpenAI | Proprietary | |
| 4 |  Grok Image (20260519)xAI | 1,388 | +/-7 | 8,010 | xAI | Proprietary |
| 5 |  Nano Banana ProGoogle Deep Mind | 1,388 | +/-3 | 386,143 | Google Deep Mind | Proprietary |
| 6 | Gemini 3.1 Flash Image PreviewGoogle Deep Mind | 1,387 | +/-4 | 196,333 | Google Deep Mind | Proprietary |
| 7 | Nano Banana ProGoogle Deep Mind | 1,385 | +/-3 | 518,323 | Google Deep Mind | Proprietary |
| 8 | GPT Image 1OpenAI | 1,373 | +/-3 | 413,464 | OpenAI | Proprietary |
| 9 | reve-2.0Reve | 1,356 | +/-11 | 2,750 | Reve | Proprietary |
| 10 | Grok Image 1.0xAI | 1,355 | +/-5 | 35,005 | xAI | Proprietary |
| 11 | uni-1.1-maxLuma AI | 1,335 | +/-5 | 33,144 | Luma AI | Proprietary |
| 12 | Grok Image 1.0xAI | 1,329 | +/-3 | 317,192 | xAI | Proprietary |
| 13 | uni-1.1Luma AI | 1,312 | +/-5 | 37,317 | Luma AI | Proprietary |
| 14 | Wan2.7 Image Pro阿里巴巴 | 1,302 | +/-4 | 40,928 | 阿里巴巴 | Proprietary |
| 15 | Seedance 2.0字节跳动Seed团队 | 1,300 | +/-3 | 753,132 | 字节跳动Seed团队 | Proprietary |
| 16 | Wan2.7 Image阿里巴巴 | 1,300 | +/-4 | 41,757 | 阿里巴巴 | Proprietary |
| 17 | HunyuanImage-3.0-Instruct腾讯AI实验室 | 1,300 | +/-4 | 227,420 | 腾讯AI实验室 | tencent-hunyuan-community |
| 18 | Nano BananaGoogle Deep Mind | 1,296 | +/-2 | 10,854,084 | Google Deep Mind | Proprietary |
| 19 | Seedream 5.0 Lite字节跳动Seed团队 | 1,293 | +/-3 | 196,437 | 字节跳动Seed团队 | Proprietary |
| 20 | Seedream 4 2K字节跳动Seed团队 | 1,271 | +/-7 | 213,012 | 字节跳动Seed团队 | Proprietary |
| 21 | Qwen-Image-2.0阿里巴巴 | 1,271 | +/-7 | 7,126 | 阿里巴巴 | Proprietary |
| 22 | FLUX 2 MaxBlack Forest Labs | 1,261 | +/-3 | 342,619 | Black Forest Labs | Proprietary |
| 23 | reve-v1.1Reve | 1,261 | +/-3 | 615,719 | Reve | Proprietary |
| 24 | Qwen-Image-2.0阿里巴巴 | 1,256 | +/-4 | 132,132 | 阿里巴巴 | Proprietary |
| 25 | OpenAI o1OpenAI | 1,250 | +/-4 | 125,175 | OpenAI | Proprietary |
| 26 | FLUX 2 ProBlack Forest Labs | 1,242 | +/-3 | 359,346 | Black Forest Labs | Proprietary |
| 27 | Qwen-Image-Edit阿里巴巴 | 1,240 | +/-3 | 1,981,564 | 阿里巴巴 | Apache 2.0 |
| 28 | reve-v1Reve | 1,234 | +/-5 | 380,861 | Reve | Proprietary |
| 29 | Qwen-Image-Edit阿里巴巴 | 1,233 | +/-3 | 332,474 | 阿里巴巴 | Apache 2.0 |
| 30 | Wan2.1-T2V-14B阿里巴巴 | 1,230 | +/-3 | 396,568 | 阿里巴巴 | Proprietary |
| 31 | FLUX 2 FlexBlack Forest Labs | 1,224 | +/-3 | 405,018 | Black Forest Labs | Proprietary |
| 32 | FLUX 2 Klein 9BBlack Forest Labs | 1,224 | +/-3 | 489,204 | Black Forest Labs | flux-non-commercial-license |
| 33 | FLUX 2 DevBlack Forest Labs | 1,220 | +/-5 | 169,269 | Black Forest Labs | flux-non-commercial-license |
| 34 | Seedream 4 High Res (FAL)字节跳动Seed团队 | 1,218 | +/-3 | 1,174,013 | 字节跳动Seed团队 | Proprietary |
| 35 | Seedream 4 (FAL)字节跳动Seed团队 | 1,210 | +/-6 | 153,917 | 字节跳动Seed团队 | Proprietary |
| 36 | p-image-editPruna | 1,210 | +/-3 | 325,602 | Pruna | Proprietary |
| 37 | reve-v1.1-fastReve | 1,206 | +/-3 | 519,059 | Reve | Proprietary |
| 38 | reve-edit-fastReve | 1,198 | +/-4 | 221,324 | Reve | Proprietary |
| 39 | FLUX 2 Klein 4BBlack Forest Labs | 1,188 | +/-3 | 489,810 | Black Forest Labs | Apache 2.0 |
| 40 | Wan2.1-T2V-14B阿里巴巴 | 1,181 | +/-3 | 371,714 | 阿里巴巴 | Proprietary |
| 41 | FLUX.1-Kontext-dev (max)Black Forest Labs | 1,181 | +/-3 | 391,727 | Black Forest Labs | Proprietary |
| 42 | FLUX.1-Kontext-devBlack Forest Labs | 1,176 | +/-3 | 6,423,947 | Black Forest Labs | Proprietary |
| 43 | FLUX.1-Kontext-devBlack Forest Labs | 1,149 | +/-3 | 3,652,096 | Black Forest Labs | flux-1-dev-non-commercial-license |
| 44 |  Step1X-EditStepFunAI | 1,139 | +/-3 | 4,950,110 | StepFunAI | Proprietary |
| 45 | GPT Image 1OpenAI | 1,138 | +/-3 | 2,861,187 | OpenAI | Proprietary |
| 46 | GPT-Image-1-miniOpenAI | 1,124 | +/-3 | 644,601 | OpenAI | Proprietary |
| 47 | Gemini 2.5 Flash Image PreviewGoogle Deep Mind | 1,081 | +/-2 | 4,965,624 | Google Deep Mind | Proprietary |
| 48 | BAGEL字节跳动Seed团队 | 1,026 | +/-6 | 13,560 | 字节跳动Seed团队 | Apache 2.0 |
| 49 | Step1X-EditStepFunAI | 998 | +/-4 | 155,733 | StepFunAI | Apache 2.0 |
Data is for reference only. Official sources are authoritative. Click model names to view DataLearner model profiles.
2026-06 Market Signals
Current Best (SOTA)
ChatGPT Image High-Fidelity
Gemini 3 Pro Image Preview 2K
Gemini 3 Pro Image Preview
Best China Model
HunyuanImage-3.0-Instruct
Seedream-4.5
Seedream-5.0-Lite
Best Open Model
- •Qwen-Image-Edit-2511
- •Qwen-Image-Edit
- •Flux-2-Klein-4B
FAQ
What is Image Edit Arena?
Image Edit Arena is a crowdsourced open evaluation platform for AI image editing models. It collects real user edit instructions, asks different models to produce edits, and uses human preference votes to rank practical editing quality.
What is the difference between inpainting and outpainting?
Inpainting edits or repairs content inside an image, such as removing an object or changing clothing.
Outpainting extends beyond the image boundary, for example expanding a square image into a wide landscape while generating matching background details.
What use cases are AI image editing models good for?
Common use cases include e-commerce product imagery, social media assets, design prototyping, photo retouching, and creative content production.
How should I choose an AI image editing tool?
Consider the editing scenario, budget, language requirements, latency, privacy, and deployment model. API models are convenient for production workflows, while open models can lower costs and support local or private deployments.