Is Qwen free? The app is, but not for commercial use

In brief

Yes, Qwen's chat app is free as of November 2026, and Qwen publishes no free message cap. Its terms allow private, non-commercial use only. New API accounts get about 1 million free tokens per model for 90 days, OpenRouter lists none of its 43 Qwen models as free, and Qwen3.8 27B's weights are Apache 2.0.

Is Qwen free to use? Yes, the chat app

Alibaba's free 1M tokens are worth at most $0.47 to $6 a model
$0.50$1$2$5Qwen3.8 FlashQwen3.7 PlusQwen3.8 Max
A violet glass knot on a black plinth beside a brushed silver padlock standing open and a chrome hourglass, in a dark studio.

You can use Qwen for nothing. Qwen's own home page calls the assistant "free to use, open to all", and the app runs at chat.qwen.ai and in the iOS, Android, macOS and Windows downloads. Qwen doesn't publish a message count for the free app on any page we could read, so there's no daily number to quote. The terms of service do say the service has paid plans, each with "a usage allowance for each subscription period". The free one just isn't named.

The chart above shows the other kind of free. It prices Alibaba's free API quota at list output rates, which is the most that 1 million tokens could be worth on each model. It's a trial of a few dollars, not a plan.

Here is every way to use Qwen without paying, from each source read on 14 November 2026:

RouteFree?CapCatch
Qwen chat app (chat.qwen.ai)YesNone publishedPrivate, non-commercial use only; you must be 18
Alibaba Model Studio API, Singapore1M tokens per model90 days from activationExpired quota is void; default setting bills on after it
OpenRouterNo Qwen model is freeNot applicable14 free models exist, all from other labs
Qwen Code with a Qwen subscriptionNoSubscription allowanceThe free sign in ended in April 2026, per a third party
Open weights, Qwen3.8 27BYes, Apache 2.0Your hardwareYou pay for the machine and the power

Is Qwen free for commercial use?

No, not in the app. The Qwen terms of service, effective 10 September 2026, state: "The Services are for private and non-commercial use only." The same agreement sets the minimum age at 18, bans using the output to train a competing model, and bans extracting data "through automated or programmatic means". A freelancer who drafts client work in the free app is, on a plain reading, outside those terms.

The terms are friendlier on two points. Qwen assigns you "any rights, title, and interest, if any" in the output. And it lets you opt out of training: "You may opt out of model training by notifying us at privacy@qwen.com." Feedback you give with a thumbs up or down is still used for training after you opt out.

Paid work has two clean routes. One is the API, where you pay per token under Alibaba Cloud's own terms. The other is open weights: Hugging Face lists Qwen3.8 27B under Apache 2.0, which lets a business run it. Licences are set model by model, so read the card for the exact one before you build on any other Qwen model.

Is the Qwen API free?

The Qwen API is free only as a trial. Alibaba Cloud's free quota page gives each new Model Studio user a quota per model, "typically 1,000,000 tokens", and says "The free quota is valid for 90 days, starting from the date you activate Alibaba Cloud Model Studio." It applies to the Singapore region only, and to real-time calls only. Batch calls, fine tuning and deployment aren't covered.

The setting that matters is "Free Quota Only". It is off by default. With it off, your calls keep working after the quota runs out and Alibaba bills the extra as pay as you go. With it on, calls stop and return the error AllocationQuota.FreeTierOnly. Turn it on if you want a hard stop. And an expired quota is void "regardless of whether it has been fully used".

Qwen API, Singapore, per 1M tokens, Alibaba list price read 14 November 2026InputOutputFree quota
qwen3.8-max$2$61M tokens, 90 days
qwen3.8-flash$0.15$0.471M tokens, 90 days
qwen3.7-plus, up to 256K input tokens$0.40$1.601M tokens, 90 days

Alibaba's price list marks the Qwen3.7 Plus rate as a limited time 20% off price, so it may change. It also sets prices by region, and the China (Beijing) list is separate. Alibaba removed older models from the new user quota on 29 October 2026, including qwen-max and qwen-plus, so a guide that names them is out of date.

Is Qwen free on OpenRouter?

No Qwen model is free on OpenRouter right now. We read its public model list on 14 November 2026. It holds 471 models, 43 of them from Qwen, and 14 carry the :free suffix. None of the 14 is a Qwen model. They come from labs such as NVIDIA, Google (Gemma), Baidu and Cohere. Older guides that point to a free Qwen coder model there are out of date.

Qwen is cheap on OpenRouter instead. Its list shows Qwen3.7 Flash at $0.03 per million input tokens and $0.13 per million output, Qwen3.8 Flash at $0.15 and $0.47, and Qwen3.8 Max at $2 and $6. Those match Alibaba's own rates for the two models Alibaba prices. If you want the free ones from other labs, Is OpenRouter free lists them with the daily request limits.

OpenRouter's limits page caps any free model at 20 requests a minute. Below 10 credits purchased you get 50 requests a day, and at 10 or more you get 1,000. That applies to the free models that exist, not to Qwen.

Is Qwen Code free?

The Qwen Code program is free, and its hosted model access is not. Qwen Code is an open source coding agent on GitHub, so the tool costs nothing to install. The README now tells you to run /auth and choose "Qwen Account with Subscription to sign in with your Plus or Pro subscription", or to bring an API key. InventiveHQ reports that Qwen's free sign in tier ended on 15 April 2026, quoting a README line that the current README no longer carries, so treat that date as a third party report.

Alibaba's own coding plan is the clearest price. Its Model Studio page, updated 4 November 2026, lists one plan, Pro, at $50 a month. It allows up to 6,000 requests per 5 hours, 45,000 per week and 90,000 per month. The cheaper Lite plan closed to new subscribers on 20 March 2026. The page names no free tier.

For a free route, run a smaller Qwen model on your own machine, or spend the 90 day API quota on a few agent sessions. A coding agent re-reads its context on every step, so a million tokens goes quicker than it sounds.

Can you run Qwen for free on your own computer?

Yes, and it's the one route with no time limit. Qwen3.8 27B is on Hugging Face under Apache 2.0, and Hugging Face's API showed about 6.8 million downloads in its 30 day count on 14 November 2026. The weights cost nothing. The electricity and the graphics card don't.

That makes local Qwen the answer to two problems at once. Your chats never leave the machine, and the commercial use clause in the app terms doesn't apply to weights you run yourself under their own licence. The catch is quality and speed: a 27B model on a home machine is slower and weaker than Qwen3.8 Max through the API. The open source LLMs guide covers what to run it with.

Who is free Qwen enough for, and when should you pay?

Free Qwen is enough for private use: study, hobby coding, drafting your own writing, trying a model. It stops being enough in two cases. One is work, because the terms bar it. The other is volume, because Qwen publishes no free cap, so you find the wall by hitting it.

Qwen's own paid chat plans are hard to price. The terms describe tiers with allowances but give no amounts, and we found no price on a Qwen page we could read. The one paid Qwen price on an Alibaba page is the $50 coding plan above.

Whizi, which publishes this page, is a different trade. Its free plan gives 10 messages a day on Whizi AI, Auto and Claude Haiku 5.5, 2 pictures a month and 5 photo uploads a day, with no card, and it has no Qwen model on it. Pro is $29.99 a month, its first week is $1.99, and it carries Claude, GPT and Gemini beside Qwen3.7 Plus at 1 credit a message. The bigger Qwen models need Powerhouse (Qwen in Whizi). For Qwen alone and for private use, Qwen's own free app wins on price.

Workflow checklist
  • Use chat.qwen.ai for private projects only, since the terms bar commercial use
  • Email privacy@qwen.com to opt out of training if you chat about anything sensitive
  • Turn on Free Quota Only in Model Studio before testing the API, or the default bills you after the quota
  • Note your Model Studio activation date, because the 1M token quota is void after 90 days
  • Pick a model for local use under Apache 2.0, such as Qwen3.8 27B, when the work is commercial
Common questions

Frequently asked questions

Is Qwen free and unlimited?

No. Qwen's app is free, but its terms describe usage allowances and say Qwen may adjust them, and it publishes no free message count. So the limit exists and its size is unstated. The API quota is a fixed 1 million tokens per model for 90 days. AI with no message limit lists tools that state their caps.

Is Qwen free for coding?

Partly. Qwen Code, the agent, is free open source software, but model access through it needs a Qwen subscription or an API key. Alibaba's coding plan costs $50 a month as of November 2026. The free paths are the 90 day API quota and local models such as Qwen3.8 27B.

Is Qwen free on OpenRouter?

No. OpenRouter's model list on 14 November 2026 had 43 Qwen models and none with a free variant. Qwen3.7 Flash is the cheapest at $0.03 input and $0.13 output per million tokens. Free models there are capped at 50 requests a day, or 1,000 after you buy 10 credits.

Is Qwen better than ChatGPT if both are free?

Qwen does not say which model its free app runs, so no score settles it. In October 2026 the Artificial Analysis Intelligence Index put Qwen3.8 Flash Next at 40 and GPT-6 Luna, the model OpenAI lists on free ChatGPT, at 38. Qwen vs ChatGPT has the full numbers, and Qwen vs DeepSeek compares the two Chinese labs.

Is Qwen open source?

Partly. Qwen3.8 27B is open weights under Apache 2.0 on Hugging Face as of November 2026, so you can download and run it for nothing. Licences are set per model card, and the chat app itself is a hosted service with its own terms, not something you can self host.