Use a Qwen Code AI Agent Free of Charge: Options and Limits

Free access to Qwen Code was shut down on April 15, 2026. Some old guides still promise two thousand free requests a day—people follow one and hit a rejection at the very first step.

Below is what works as of August 28, 2026: what is free in Qwen Code and will remain free, what costs money, which free options are still available, and which regional restrictions to keep in mind.

Ailoot documentation: API, keys, and setup.

Is Qwen Code really free?#

Qwen Code is an AI agent from Alibaba that runs in the terminal. It reads your project, edits files, and runs commands. The official description in the GitHub repository is “an open-source AI agent for coding that lives in your terminal.”

The code is open-source, under the Apache 2.0 license. You install it with a single line through npm; all you need is Node.js 22 or later. Nobody ever pays for the installation.

The costs start when the agent goes to the model. Every time Qwen Code thinks about your task, it sends a request to Alibaba's server—and access to that model is sold separately.

So the question “Is Qwen Code paid or free?” splits in two. The program is free, but the brains it uses to think are another matter, and that's what this article is about.

What happened to free access to Qwen Code?#

It used to be simple. You installed Qwen Code, and the first time you ran it, it opened a sign-in page. You logged in with your qwen.ai account, and then the agent worked. No card, no key. This method was called Qwen OAuth.

They removed it. The Qwen Code authentication documentation has a direct statement: the free Qwen OAuth plan was canceled as of April 15, 2026. It also clarifies that old saved sign-ins might keep working for a little while, but the server rejects new requests.

The “sign in with qwen.ai” option was also removed from the authentication menu. Previously, the /auth command inside Qwen Code offered a free sign-in option. Now it's no longer on the list at all—only paid options and connecting an external key remain.

Why do guides still promise 2,000 requests a day?#

Two thousand requests a day was a real figure under the old setup. While free access worked, that was the daily allowance, and it was enough for personal projects. In the space of a few months, everyone had written about it—from blogs to YouTube videos.

Materials don't disappear when a plan changes. They stay up with their old dates, look current, and describe Qwen as it was four months ago. You open one and follow the steps—and get nowhere when you try to sign in through qwen.ai, because the option is gone.

Here's a rough way to check whether a guide is current: if it promises free access through a qwen.ai account or “2000 requests a day” without a single word about the April cancellation, it's outdated. There's no point following it further.

This is a common problem with fast-moving tools. Model plans are rewritten every couple of months, while search results live on for years. So the first thing to check in any guide is its publication date and the date it was last verified—they tell you which month it is talking about.

What is still free in Qwen Code in 2026?#

After April, free options didn't disappear—they just stopped being a one-click setup. Here's what is actually available for free at the end of August 2026 and what you'll have to give up in exchange.

Next, we'll look at each option separately: what it offers, where it trips up, and whether it's right for you. Qwen Code subscriptions aren't covered here—there's a separate breakdown of how to pay for Qwen Code.

How is Qwen browser chat different from Qwen Code in the terminal?#

“But Qwen works for me for free”—that's what people say about browser chat. And it's true: the Qwen website has a chat where you can type, just like any other chat with a model. It doesn't ask you for money.

But that's not Qwen Code. The chat responds with text and code that you copy over by hand. It can't see your project, open files, or run anything itself. You're the courier between the chat and your computer.

Qwen Code works differently. It runs in the terminal, reads the entire project folder, edits the files it needs, and runs commands itself. You don't need to shuttle code back and forth—it works right inside your project.

That's the difference you pay for. The chat answers a single question, and Alibaba provides it for free. An agent that digs around in a project uses far more model requests, and access to that is paid.

Here's the practical takeaway. If you need to ask a question and get a piece of code, free browser chat is enough. If you need an assistant that can inspect a project and make changes on its own, that's Qwen Code, and it won't be free.

Can you run Qwen Code for free on your own computer?#

Qwen Code has a mode that's easy to overlook when you're in a hurry: working with a local model. The documentation calls it Custom Provider: connecting your own server or a local engine such as Ollama and vLLM. You install the model on your computer, and the agent no longer goes to Alibaba's cloud. All requests stay on your own computer.

The cost in money is zero. Open Qwen models are free to download and run at home, and every agent request stays on your machine. No card, no key, no request limits. For a breakdown of which open models can handle coding at all, see free models for coding.

The other cost is hardware. A model that writes code well takes a lot of graphics memory. This is doable on a gaming computer with a powerful graphics card. On a regular laptop, the model either won't fit, or it'll take so long to think over each answer that paying would be easier.

There's also a second cost: the time you'll spend setting it up. Installing a local model and connecting it to Qwen Code through the settings takes an evening of following instructions and tinkering; it isn't a one-click setup. If you like keeping everything to yourself and not sending data elsewhere, it's worth it. If you just need to get a project done quickly, probably not.

Your data also stays put—that's another argument for running a model locally. The code stays on your computer, and nothing is sent to someone else's server. If the project is sensitive, that matters more than the savings.

Are free quotas from third-party providers a way out?#

Qwen Code has supported different model providers from the beginning, not just Alibaba. The official provider list includes DeepSeek, Grok, MiniMax, Kimi, ModelScope, and several others. You connect to them in the same way: get a key from the provider and enter it in the Qwen Code settings.

Some providers offer free access with restrictions. One might give you a few free requests a day; another might make specific models available for free. This is what remains in place of sign-in through qwen.ai, except it isn't built in—you have to set it up yourself.

There are two catches. First, setup: you have to register with the provider, create a key, and enter it carefully in the Qwen Code settings. That's noticeably more complicated than “sign in with one click,” as it used to be. Second, the free allowance: it's enough to try things out and handle small tasks, but usually not enough for a large project.

There are no exact figures for free quotas here: each provider has its own, and they change often. Check the provider's own page when you connect—any number in an article will be outdated by then.

So what should you choose for your task?#

It's all in one table—choose what suits your situation.

In short: free Qwen Code now costs either time spent on setup or a powerful computer. The old “sign in and it all works for free” option was shut down on April 15, 2026, and guides that still promise it simply haven't been updated.

If you decide to pay for access after trying the free options, there's a separate breakdown of how to pay for Qwen Code, why your card might not go through, and where money can leak outside the subscription. And if you'd like an option with no payment and no personal hardware, see which free AI agent to install in VS Code.

If you need access to neural networks through Ailoot, take a look at Ailoot.