Skip to content

DirectoryTools

Qwen-Image-3.0 logo

Qwen-Image-3.0

Alibaba's hosted image model for dense layouts, small text and multilingual images.

Visit Qwen-Image-3.0

Image & DesignFreemiumWe checked the prices on alibabacloud.com on 28 Sep 2026.

Qwen-Image-3.0 homepage screenshot

Recaply review

Qwen-Image-3.0 review: cheap text-heavy images, no open weights

Qwen-Image-3.0 is Alibaba's third-generation Qwen image model, built for infographics, slides and interface mockups packed with small text, and sold per image through Alibaba Cloud Model Studio and Qwen Cloud.

4.5K
Tokens of prompt it can follow, up from 1K on Qwen-Image-2.0
10px
The smallest text the Qwen team shows staying legible
12
Languages it renders natively, including Chinese, English, Japanese and Korean
$0.03
Per image on Qwen Cloud; $0.024754 in Alibaba Cloud's US region

What is Qwen-Image-3.0?

Qwen-Image-3.0 turns a prompt of up to about 4,500 tokens, plus up to three reference images, into one to six images per request, so a single call can lay out a newspaper page, a storyboard grid or app screens nested inside each other. The same model makes and edits images. It comes in two versions: the standard model and Pro, the flagship, which is tuned for text that blends into real surfaces and costs more per image.

It is Alibaba's current top image model. Unlike the first Qwen-Image releases, its weights are closed: it runs only on Alibaba's APIs, and the newest open Qwen image model is Qwen-Image-2.1, a separate 7B release.

What changed from Qwen-Image-2.0?

Mostly prompt length and price: it follows much longer prompts and costs less per image, but the free allowance is smaller and the weights stay closed.

Prompt length
Up to about 4,500 tokens, against about 1,000 for Qwen-Image-2.0.
Price
In the Singapore region the standard model costs $0.03 per image against $0.035 for 2.0, and Pro costs $0.04 at 1K against $0.075 for 2.0 Pro.
Prompt help
Prompt rewriting gains an agent mode for text-to-image, and a thinking step runs by default; both help short prompts but add time.
Free images
10 free images per model in Singapore, down from 100 for the 2.0 models.

Who is Qwen-Image-3.0 best for?

Qwen-Image-3.0 is best for teams that make information-dense images with lots of small text, such as infographics, slides and storyboards, in several languages and at API volume.

  • Dense infographics and slides

    Prompts can run to about 4,500 tokens, so one image can hold a full grid of labeled panels, formulas and charts.

  • Small print that must stay readable

    The Qwen team shows text as small as 10 pixels rendered legibly, useful for newspapers, paper pages and product labels.

  • Multilingual layouts

    It renders text natively in 12 languages, including Chinese, English, Japanese, Korean and Spanish.

  • High-volume API jobs

    At $0.024754 per standard image in the US region, it costs about half as much as Seedream 5.0 Pro at 1.5K and far less than Nano Banana Pro.

What's included with Qwen-Image-3.0?

Making and editing in one model, prompt rewriting with a thinking step, sizes up to 2048x2048 in ratios up to 8:1, six images per call and an OpenAI-compatible API.

One model for making and editing
Text-to-image and editing share one endpoint, and an edit can take up to three reference images.
Prompt rewriting and thinking
Both are on by default to lift short prompts, with an agent-style rewriter for text-to-image; turn them off for exact control or faster results.
Flexible sizes
Any size from 512x512 to 2048x2048 in aspect ratios from 1:8 to 8:1, or let the model choose.
Up to six images per call
Ask for one to six variations in a single request.
OpenAI-compatible endpoint
Generation and editing work through OpenAI's images format; asynchronous batch jobs need Alibaba's DashScope API.
Live web knowledge
It can look up current information online, such as a city's weather forecast, and draw it into the image.

How much does Qwen-Image-3.0 cost?

Qwen-Image-3.0 is billed per image: $0.024754 at 1K or 2K in Alibaba Cloud's US region, or $0.034380 at 1K and $0.068761 at 2K for the Pro version.

qwen-image-3.0$0.024754Per image, 1K or 2K
  • $0.002750 per input image for edits
  • Text-to-image and image editing in one model
  • Sizes from 512x512 to 2048x2048, ratios 1:8 to 8:1
qwen-image-3.0-pro$0.034380Per 1K image
  • $0.068761 per 2K image
  • $0.002750 per input image for edits
  • Flagship version, tuned for text that blends into real materials

Prices are for Alibaba Cloud Model Studio's US (Virginia) region; Beijing, Hong Kong, Frankfurt and Tokyo charge the same. Singapore and Qwen Cloud charge $0.03 per image for the standard model, $0.04 (1K) or $0.075 (2K) for Pro and $0.003 per input image, and give 10 free images per model, valid for 90 days. Failed generations are not billed. Qwen Cloud's Token Plan, from $8 a month for 11,500 credits, also covers the Pro model, but Qwen does not publish how many credits an image uses. Full pricing on alibabacloud.com. We checked these prices on .

How do readers rate Qwen-Image-3.0?

No one has rated Qwen-Image-3.0 yet. Be the first.

No ratings yetThe score and its stars show once 3 readers have rated Qwen-Image-3.0.
Your rating

Sign in to rate. One rating per account.

Your rating of Qwen-Image-3.0

Qwen-Image-3.0 pros and cons

It handles long, text-heavy layouts for less than its rivals, but the weights are closed and its text still needs proofreading.

Pros

  • Testers say it handles long layout prompts and small text better than most image models, which is where Alibaba focused this version.
  • Costs less than its closest rivals: $0.024754 per standard image in the US region, against $0.045 for Seedream 5.0 Pro and $0.134 for Nano Banana Pro.
  • Users call it a clear step up from Qwen-Image-2.0, with better backgrounds, and it follows prompts of about 4,500 tokens against 2.0's 1,000.
  • Generation and editing share one API with an OpenAI-compatible endpoint, so it fits into existing image pipelines.

Cons

  • Closed weights: it runs only on Alibaba's APIs, and many users who ran earlier Qwen image models locally say that rules it out.
  • Text is readable but not always right: users found misspelled Korean, a dropped dash and made-up chart data, so proofread anything you publish.
  • Several testers rate it behind GPT Image 2 and Nano Banana Pro for general image quality and UI designs, and some spotted anatomy errors such as wrong toe counts or an extra leg.
  • Some users say its filters refuse harmless prompts, and the thinking step Alibaba turns on by default adds generation time.

What are the best Qwen-Image-3.0 alternatives?

Four tools on Recaply do Qwen-Image-3.0's job. GLM-Image and ERNIE-Image are open source, so you can run them on your own server.

Questions about Qwen-Image-3.0

What is the difference between Qwen-Image-3.0 and Qwen-Image-3.0 Pro?

Pro is the flagship version, tuned to render text that blends naturally into physical materials. It costs $0.034380 at 1K and $0.068761 at 2K in the US region, against a flat $0.024754 for the standard model, and Qwen Cloud allows it 5 requests a minute instead of 20.

Who is Qwen-Image-3.0 for?

Qwen-Image-3.0 is best for teams that make information-dense images with lots of small text, such as infographics, slides and storyboards, in several languages and at API volume.

Is Qwen-Image-3.0 free?

Only as a short trial. Alibaba Cloud Model Studio's Singapore region and Qwen Cloud give 10 free images per model, valid for 90 days. After that, and in every other region, each image is billed, from $0.024754 per image in the US region.

How much does Qwen-Image-3.0 cost?

Qwen-Image-3.0's paid plans are qwen-image-3.0 at $0.024754 per image, 1K or 2K and qwen-image-3.0-pro at $0.034380 per 1K image.

Is there an open source alternative to Qwen-Image-3.0?

Yes. GLM-Image and ERNIE-Image are open source tools that do the same job, so you can run your own.

What are the best Qwen-Image-3.0 alternatives?

The closest alternatives on Recaply are Seedream 5.0 Pro, Nano Banana Pro, GLM-Image and ERNIE-Image.

Can I download Qwen-Image-3.0 and run it myself?

No. Qwen-Image-3.0 has closed weights and runs only through Alibaba Cloud Model Studio and Qwen Cloud. The newest open Qwen image model is Qwen-Image-2.1, a separate 7B generation and editing model released under the Qwen Research License.

Does Qwen-Image-3.0 work with the OpenAI SDK?

Yes. Qwen-Image-3.0 has an OpenAI-compatible images/generations endpoint for both making and editing images. It does not support OpenAI's /images/edits route, masks, streaming or asynchronous calls there, so use Alibaba's DashScope API for batch jobs.

How long are Qwen-Image-3.0 image links valid?

Generated images come back as PNG links that expire after 24 hours, when Alibaba also deletes the task data, so download and store every image you want to keep straight away.

Our verdict

Qwen-Image-3.0 is a good buy for teams that generate text-heavy images through an API and watch the cost.

It follows far longer layout prompts than Qwen-Image-2.0 and costs less per image than Seedream 5.0 Pro or Nano Banana Pro, though testers still rate Nano Banana Pro higher for general image quality.

Skip it for
Anyone who needs open weights to run locally, 4K output, or text that is right without proofreading.
Check first
Run your longest real prompts, in every language you publish in, through the free trial images before you commit.