Images · Alibaba (Qwen)

Qwen Image

Alibaba's models for images full of text: posters, menus, infographics. Qwen Image 3 stays the pick; 2.1 Turbo is the cheapest.

Anonymized Low cost
Latest versionQwen-Image 2.1 TurboOctober 9, 2026
Artificial Analysis Text to Image Arena1,090 Elo#11 of 37 on Artificial Analysis Text to Image Arena
Price$0.018 per imageQwen Image 3

What sets it apart

Alibaba's family for images full of text: posters, menus, infographics. The first Qwen-Image was an open 20B model with strong Chinese and English lettering; version 3 is closed, supports 12 languages and prompts up to 4,500 tokens, and version 2.1 is a 7B model whose Turbo weights are open for research only.

  • Dense text layouts in many languages
  • v1 open weights, Apache 2.0; 2.1 research licence
  • v3 closed, 12 languages announced

Recommended for: put readable words in the picture: a poster, a label, a menu.

Strengths and limits

Where it shines

  • Dense layouts: newspapers, menus, storyboards, interfaces
  • Small text in many languages (12 announced for version 3)
  • Close prompt adherence at about 2 credits

Think twice when

  • Anonymized, not private: Alibaba Cloud's policy applies
  • Seven versions in the catalog: apart from 2.1 Turbo on price, the older ones are rarely the better buy
  • Qwen Image 3 and 3 Pro: 1K or 2K output
  • Prompts up to 10,000 characters (2 and 3)

Use cases

A restaurant menu

A one-page bistro menu with three sections: Starters, Mains, Desserts, each with four dishes and prices in euros, elegant serif type, cream paper.

Dense layouts with small text are its specialty.

A bilingual poster

A tea shop poster with the English title 'Morning Leaf' and a Chinese subtitle below it, ink-wash illustration of tea mountains.

It was built for Chinese and English lettering.

Against its rivals

  • IdeogramIdeogram treats text as part of the design; Qwen fits denser layouts such as menus and interfaces.#23 of 37 on Artificial Analysis Text to Image Arena (1,013 Elo)
  • GPT ImageGPT Image ranks higher in the arena; Qwen is built for denser text layouts such as menus and interfaces.#1 of 37 on Artificial Analysis Text to Image Arena (1,197 Elo)
  • SeedreamSeedream is stronger on a consistent subject across a series; Qwen on dense text.#12 of 37 on Artificial Analysis Text to Image Arena (1,081 Elo)
Compare side by side ↗

Version history

Each version, dated, with what it changed over the one before.

Artificial Analysis Text to Image Arena, version after versionElo rating · Artificial Analysis

Hollow points are earlier versions no longer offered in Sub Rosa.

See the numbers
VersionReleasedElo
Qwen-Image2025-08-04889
Qwen-Image 2.02026-02-10962
Qwen-Image 2.0 Pro2026-04-221,031
Qwen-Image 3.0 and 3.0 Pro2026-07-211,090
Qwen-Image 2.1 Pro2026-10-021,054
Qwen-Image 2.1 Turbo2026-10-091,028
  1. Qwen-Image
    • Open 20B model with strong complex text rendering, especially Chinese
    Source : huggingface.co
  2. Qwen-Image 2.0
    • Second generation announced with its own technical report
    Source : stdaily.com
  3. Qwen-Image 2.0 Pro
    • Pro tier listed on the Artificial Analysis leaderboard on 2026-04-22
    Source : artificialanalysis.ai
  4. Qwen-Image 3.0 and 3.0 Pro
    • Closed model with 4,500-token prompts and 12 natively displayed languages
    • No weights released, a break from the earlier versions, which were open under Apache 2.0
    Source : drweb.de
  5. Qwen-Image 2.1Earlier, not in Sub Rosa
    • Unified 7B text-to-image and editing model with native transparent (RGBA) output, up to 10 reference images and local edits by circle, annotation or mask
    • Open weights on Hugging Face under the Qwen Research License (non-commercial), a break from the earlier Apache 2.0 releases
    Source : orcarouter.ai
  6. Qwen-Image 2.1 Pro
    • Hosted Pro tier through Qwen Cloud and Alibaba Cloud Model Studio, up to 2048 by 2048 pixels, aspect ratios from 1:8 to 8:1
    Source : llmreference.com
  7. Qwen-Image 2.1 TurboLatest
    • Accelerated checkpoint that renders in 8 denoising steps instead of 40, same 7B architecture
    Source : marktechpost.com

Benchmarks

How to read these scores: what a benchmark tells you.

Artificial Analysis Text to Image Arena: #11 of 37 models measured in the catalog. Thousands of people compare two images made from the same prompt without knowing the models, and pick the better one. An Elo rating, as in chess: higher is preferred more often. 100 points apart means the higher one wins about 64% of duels.

Artificial Analysis Text to Image ArenaElo rating · Artificial Analysis, October 4, 2026
  1. 1. GPT Image 2.5 SunburstGPT Image1,197
  2. 2. GPT Image 2.5 FlareGPT Image1,191
  3. 3. GPT Image 2GPT Image1,172
  4. 4. Nano Banana 2.1Nano Banana1,172
  5. 5. Grok Imagine 2.0Grok Imagine1,155
  6. 6. Nano Banana 2Nano Banana1,125
  7. 7. Muse ImageMuse Image1,115
  8. 8. GPT Image 1.5GPT Image1,107
  9. 9. Nano Banana ProNano Banana1,102
  10. 10. Nano Banana 2 LiteNano Banana1,096
  11. 11. Qwen Image 3 ProQwen Image1,090
  12. 13. Qwen Image 3Qwen Image1,076
  13. 14. Qwen Image 2.1 ProQwen Image1,054
See the numbers
ModelElo
1. GPT Image 2.5 Sunburst1,197
2. GPT Image 2.5 Flare1,191
3. GPT Image 21,172
4. Nano Banana 2.11,172
5. Grok Imagine 2.01,155
6. Nano Banana 21,125
7. Muse Image1,115
8. GPT Image 1.51,107
9. Nano Banana Pro1,102
10. Nano Banana 2 Lite1,096
11. Qwen Image 3 Pro1,090
13. Qwen Image 31,076
14. Qwen Image 2.1 Pro1,054

All 7 models in this family

ModelUsePrivacyPrice
Qwen Imageqwen-imageImageAnonymized$0.0135 per image
Qwen Image 2qwen-image-2ImageAnonymized$0.0225 per image
Qwen Image 2 Proqwen-image-2-proImageAnonymized$0.045 per image
Qwen Image 3qwen-image-3ImageAnonymized$0.018 per image
Qwen Image 3 Proqwen-image-3-proImageAnonymized$0.0225 per image
Qwen Image 2.1 Proqwen-image-2-1-proImageAnonymized$0.0225 per image
Qwen Image 2.1 Turboqwen-image-2-1-turboImageAnonymized$0.009 per image

Specifications

SpecificationQwen-Image 3 and 3 ProQwen-Image 2 and 2 ProQwen-ImageQwen-Image 2.1 TurboQwen-Image 2.1 Pro
ReleasedJuly 21, 2026February 10, 2026August 4, 2025October 9, 2026October 2, 2026
ParametersNot disclosedNot disclosed20B7BNot disclosed
Weights··Open, Apache 2.0Open, Qwen Research License (non-commercial)·
Acceptstexttexttexttext, imagestext, images
Producesimagesimagesimagesimagesimages
Resolutions1K, 2K··1K, 2K1K, 2K
Price$0.018 per image$0.0225 per image$0.0135 per image$0.009 per image$0.0225 per image
PrivacyAnonymizedAnonymizedAnonymizedAnonymizedAnonymized

Prices, context and lengths from the live catalog on October 16, 2026; parameters and licenses from the maker.