Best AI Image Generators: Text-to-Image Leaderboard
Ranked by LMArena scores from anonymous head-to-head human votes. Artificial Analysis Image Arena Elo is shown for comparison; the two scales use different baselines and cannot be compared directly.
- Top open-weight
Qwen Image 2.11223Alibaba - #1 on AA (matched)GPT Image 2.5 SunburstAA Elo 1222OpenAI
On LMArena’s text-to-image leaderboard published Oct 7, 2026, #1 is GPT Image 2.5 Sunburst by OpenAI (1425, 17,752 votes); the image-editing leader is GPT-Image-2.5 Sunburst (1514).
Full leaderboard · Text-to-image
19 modelsSorted by LMArena score by default; ± is half the 95% confidence interval.
| # | Model | LMArena | Artificial Analysis | Released | License | |||
|---|---|---|---|---|---|---|---|---|
| Score↓ | Rank | Votes | Elo | Rank | ||||
| 1 | Qwen Image 2.1AlibabaOpen |
1223±7 | 19 | 11,441 | 1042±12 | 22 | Sep 2026 | qwen-research |
| 2 |  Ideogram 4.0 QualityIdeogramOpen |
1205±4 | 20 | 60,591 | – | – | – | Ideogram Open Model |
| 3 |  Cosmos3-Super-Text2Image (Agentic)NVIDIAOpen |
1165±7 | 27 | 10,805 | – | – | – | OpenMDW-1.1 |
| 4 | Cosmos3-Super-Text2ImageNVIDIAOpen |
1152±6 | 33 | 18,485 | – | – | – | OpenMDW-1.1 |
| 5 |  Hunyuan Image 3.0TencentOpen |
1150±3 | 35 | 180,518 | 989±16 | 54 | Sep 2025 | tencent-hunyuan-community |
| 6 |  FLUX 2 DevBlack Forest LabsOpen |
1146±4 | 38 | 77,988 | – | – | – | flux-non-commercial-license |
| 7 | Qwen Image 2512AlibabaOpen |
1125±3 | 44 | 98,288 | – | – | – | Apache 2.0 |
| 8 | HHiDream O1 ImageHiDreamOpen | 1116±4 | 47 | 66,906 | 975±10 | 63 | May 2026 | MIT |
| 9 |  Krea 2 TurboKreaOpen |
1113±4 | 52 | 51,552 | – | – | – | krea-2-community-license |
| 10 | Z-Image TurboAlibabaOpen |
1083±5 | 59 | 24,808 | 936±11 | 72 | Dec 2025 | Apache 2.0 |
| 11 | FLUX 2 Klein 9BBlack Forest LabsOpen |
1070±3 | 62 | 150,782 | – | – | – | flux-non-commercial-license |
| 12 | Qwen Image Prompt ExtendAlibabaOpen |
1060±3 | 63 | 716,773 | – | – | – | Apache 2.0 |
| 13 | Qwen ImageAlibabaOpen |
1057±3 | 66 | 84,550 | 903±18 | 83 | Aug 2025 | Apache 2.0 |
| 14 | FLUX 2 Klein 4BBlack Forest LabsOpen |
1030±3 | 70 | 152,389 | – | – | – | Apache 2.0 |
| 15 |  GLM ImageZ.aiOpen |
1012±9 | 76 | 4,854 | 885±18 | 88 | Jan 2026 | MIT |
| 16 | FLUX 1 Dev FP8Black Forest LabsOpen |
969±4 | 78 | 49,221 | – | – | – | Open |
| 17 | FLUX 1 Kontext DevBlack Forest LabsOpen |
941±4 | 80 | 216,225 | – | – | – | flux-1-dev-non-commercial-license |
| 18 | ?Stable Diffusion 3.5 LargeOpen | 938±5 | 81 | 23,396 | – | – | – | Open |
| 19 |  BAGELByteDanceOpen |
898±6 | 82 | 12,363 | 594±35 | 106 | May 2025 | Apache 2.0 |
Scores: LMArena leaderboard-dataset (lmarena-ai on Hugging Face, CC-BY-4.0), Text-to-image published Oct 7, 2026; AA columns from the Artificial Analysis Image Arena. Please credit the sources when citing.
Sources & methodology
- LMLMArena leaderboard dataset (Hugging Face, CC-BY-4.0)
published Oct 7, 2026 · 82 models
- AAArtificial Analysis Image Arena
fetched Oct 26, 2026 · 106 models
- Scores come from LMArena’s text_to_image and image_edit leaderboards (overall category, latest release): users vote between two anonymous outputs and LMArena fits a Bradley–Terry model.
- AA values are shown only when the model name (ignoring bracketed variant notes) maps to exactly one row; hover to see the AA entry name. Anything unmatched stays blank (–); nothing is guessed.
- Price, resolution and generation speed have no reliable public source yet and are not shown.
- Synced daily at 06:55 UTC; public JSON: /api/public/image-leaderboard.php
FAQ
What is the best AI image generator right now?
On LMArena’s text-to-image leaderboard published Oct 7, 2026, #1 is GPT Image 2.5 Sunburst by OpenAI (1425, 17,752 votes); the image-editing leader is GPT-Image-2.5 Sunburst (1514).
Which vendors rank highest for image generation?
On the text-to-image board, each vendor's best model ranks: OpenAI GPT Image 2.5 Sunburst (1425, #1), xAI Grok Imagine Image 2.0 (Canvas) (1334, #4), Microsoft AI MAI Image 2.6 (1332, #5), Google Gemini Nano Banana 2.1 (1328, #6), Reve Reve 2.1 (1302, #7).
Which open-weight image models can I run locally?
19 text-to-image models have non-proprietary licenses; the highest is Qwen Image 2.1 (1223, license qwen-research). Some licenses (e.g. FLUX dev non-commercial) restrict commercial use.
Why do LMArena and AA scores differ so much?
The two platforms use different voters, prompts and baselines, so absolute numbers are not comparable; relative order within one platform is what matters.