What is Qwen? Qwen3.8, free chat app and API pricing 2026

Qwen is Alibaba's family of AI models, and the newest is Qwen3.8. Where to use it free, open weights and licenses, API prices, and language support.

Adam Wang

31/10/2026 · 6 min read

Facebook

What is Qwen? Qwen3.8, free chat app and API pricing 2026

In this article (7)

Qwen is Alibaba's model family, and the newest generation is Qwen3.8
Use Qwen free at chat.qwen.ai and in the apps
Qwen3.8-27B ships under Apache 2.0, the bigger weights come with conditions
How much does the Qwen API cost?
Qwen understands Vietnamese, and the official pages name it
Pick the way you use Qwen by the job
Sources

Qwen is Alibaba's family of AI models and its chat assistant. You can use Qwen free on the web and in the apps, download open-weight versions, or pay per token through the API.

Qwen is Alibaba's model family, and the newest generation is Qwen3.8

On Aug 3, 2026, the Qwen team announced Qwen3.8-Max, which it called the most capable model in the Qwen family to date. It has 2.4 trillion parameters, with 95 billion active on each pass. It is also the first Max-class Qwen model with open weights.

The Qwen3.8-Max: A New Bar for Coding and Cowork blog post, dated Aug 3, 2026, on qwen.ai — The Qwen3.8-Max launch post, aimed at coding and long office tasks. Screenshot: Qwen blog, Oct 12, 2026.

Qwen3.8 is built on the Qwen3.5 architecture. After the Max release, the Qwen Research page posted Qwen3.8-Omni-Flash and Qwen3.8-LiveTranslate on Sep 18, 2026, then the Qwen-Image-2.1 image model on Sep 20, 2026. As of Oct 27, 2026, those are still the newest posts on that page.

Use Qwen free at chat.qwen.ai and in the apps

The qwen.ai homepage describes Qwen as an AI assistant that is free to use and open to all. The web version lives at chat.qwen.ai, with Log in and Sign up buttons to create an account. Many people call it Qwen Chat, while the button on the Qwen blog says Qwen Studio.

If you want to compare it with another free assistant that also offers open-weight models, see our explainer on DeepSeek.

Three cards on qwen.ai: the free Qwen assistant, the OpenAI-compatible API Platform, and Download — Three main entry points: the Qwen assistant, the API Platform, and the apps. Screenshot: qwen.ai homepage, Oct 12, 2026.

The prompt box on the homepage offers Thinking (step-by-step reasoning), Search (web search), and Deep Research (multi-step research ending in a report). Next to them are Web Dev (build a web page from a description), Image Generation, and Video Generation. Besides the web, the homepage lists apps for iOS, Android, macOS, and Windows.

The privacy policy, effective Sep 7, 2026, says the service is for users 18 and older. To compare the main chatbots, browse the news section.

Qwen3.8-27B ships under Apache 2.0, the bigger weights come with conditions

On the Qwen account on Hugging Face, the Qwen3.8 generation has three main open-weight releases, each under a different license. Check the license before using one in a paid product.

The Qwen3.8-27B model page on Hugging Face, with the License apache-2.0 tag and 28B params model size — The Qwen3.8-27B model page, with the apache-2.0 license tag right under the name. Screenshot: Qwen on Hugging Face, Oct 12, 2026.

Qwen3.8-27B is a 27 billion parameter dense model that understands images and video, released under Apache 2.0. Its native context is 262,144 tokens, extendable to 1 million. It is the easiest model in the 3.8 line to run yourself.

Qwen3.8-2.4T-A95B is the open-weight version of Qwen3.8-Max, but without the image input and non-thinking mode of the cloud model. Its license sets two conditions. A product with more than 100 million monthly active users or more than 20 million USD in monthly revenue must display the model name in its interface. A company selling a model service or AI work assistant with more than 50 million USD in revenue over 12 months needs a separate license.

Qwen3.8-Flash-Next is a preview of the architecture planned for Qwen4, under the Qwen Community 1.0 license. That license requires any company selling a model service or AI work assistant to get a separate license before commercial use, with no revenue threshold.

How much does the Qwen API cost?

A note on the Hugging Face model pages says the official Qwen API service is Qwen Cloud. The Qwen3.8-Max page on Qwen Cloud lists Alibaba Cloud Model Studio as the inference provider. The API is OpenAI-compatible, so many coding tools can call Qwen with little change.

Qwen3.8-Max pricing on Qwen Cloud: 2 USD input and 6 USD output per 1 million tokens, with a 1M context window — Per-token pricing for Qwen3.8-Max, with its 1 million token context limit. Screenshot: Qwen3.8-Max on Qwen Cloud, Oct 12, 2026.

Prices per 1 million tokens, rechecked Oct 27, 2026:

Model	Input	Output
qwen3.8-max	$2	$6
qwen3.8-flash	$0.15	$0.47

The max price is from the Qwen Cloud page; the flash price is from the Model Studio pricing page (Singapore region, International scope), which lists qwen3.8-max at the same $2 and $6. For qwen3.8-max, input that hits the implicit cache drops to $0.25.

The free quota page says new users who activate Model Studio in the Singapore region get a separate quota for each model, usually 1 million tokens, valid for 90 days. You must complete your account information first. Once the quota runs out, billing starts automatically unless you turn on Free Quota Only.

If you use Qwen through coding tools, Qwen Cloud offers the Token Plan, a subscription covering text, vision, speech, and image models, in Individual and Team versions. On Oct 27, 2026, viewed for the Singapore region, the monthly Individual plan had four discounted tiers: Lite at $6 (list $8, 11,500 credits a month), Essential at $10 (list $16), Standard at $18 (list $25), and Pro at $68 (list $80, 16 times Lite's credits). The page calls these special-offer prices, so check again before you buy. It works with Qwen Code, Claude Code, Cursor, Codex, Cline, and other tools that support the OpenAI or Anthropic protocols.

Qwen understands Vietnamese, and the official pages name it

The Qwen3.5 model page on Hugging Face says the line supports 201 languages and dialects, without listing each one. Qwen3.8 builds on Qwen3.5, and the Qwen3.8 pages do not repeat a language count.

The Supported Languages table for Qwen3.8-LiveTranslate, with Vietnamese in both the 60 input languages and the 29 voice output languages — Vietnamese appears in both language groups for LiveTranslate. Screenshot: Qwen blog, Oct 12, 2026.

The clearest mention is in the Qwen3.8-LiveTranslate post, a live speech translation model. Its language table lists Vietnamese among 60 languages for audio in and text out, and among 29 languages for spoken output. No per-language score is given, so test real sentences from your own work first.

Pick the way you use Qwen by the job

For everyday chat, open chat.qwen.ai or the app. For coding or batch jobs over the API, compare qwen3.8-flash and qwen3.8-max using the prices above. If your data must stay on your own machine, run Qwen3.8-27B yourself under Apache 2.0.

On data, the privacy policy says the Qwen assistant is run by NTH Power Global Tech Singapore, your data is processed and stored in Singapore, and AI inference is handled by Alibaba Cloud (Singapore). De-identified content may be used to train models, and the policy says your raw input is not used to train models directly. The policy also asks you not to upload ID documents, health records, or financial account details.