Is Qwen Code free: what changed and how to keep working

The question "is Qwen Code free" came up because the word "free" applied to two different things: the program itself and the model that ran in it. Nothing changed for the program, but the free login through a Qwen account is closed. Below we go through what exactly changed per the project documentation, which options remain and what working on your own key costs. Prices are as of October 18, 2026.

What is free and what is not

Two parts must be separated.

1. The Qwen Code program. It is an open-source terminal agent, its repository is on GitHub (QwenLM/qwen-code), and the license terms are stated there. You can install and run it without paying.
2. The model's work. Every request to a model consumes compute, and someone pays for it: you, your employer, or a provider who decided to hand out access for free for promotion.

Qwen Code used to offer a login through a qwen.ai account (Qwen OAuth mode) with a free allowance. According to the project documentation, that free tier was discontinued on April 15, 2026. The same page says that already saved tokens may keep working for a while, but new requests will be rejected. This login is no longer selectable in the /auth dialog. If the old Qwen account login stopped working, this is most likely why: pick another connection method in /auth.

There are no promises here to bring free access back or to get around this limit, and there cannot be. Terms change on the developers' side, so check the README and documentation for the current status.

Which options remain

From the Qwen Code authentication docs. The list changes, so check the current version.

We do not give Alibaba subscription prices and terms here: check the current plans on the service's site.

Continuing on your own Zukko key

Zukko works as a "custom provider" with an OpenAI-compatible address. A single dollar balance, pay per token, no subscription.

export OPENAI_API_KEY="fetch_your_key"
export OPENAI_BASE_URL="https://api.zukko.pro/v1"
export OPENAI_MODEL="qwen3.8-flash"
qwen

Create the key on the API keys page↗. A persistent variant via ~/.qwen/settings.json is described in Qwen Code CLI: connecting↗. The full list of models and prices: the models page in the docs↗ and GET https://api.zukko.pro/v1/models.

Access from Russia goes through the gateway. The balance is topped up from $1: SBP (the payment page adds 8.5% on top), MasterCard or cryptocurrency. With SBP, rubles are charged at the market rate at the moment of payment.

What it costs in money

Take a hypothetical small agent task (the volumes are an assumption for the example): 200K input tokens and 20K output tokens. Formula: input in millions × input price + output in millions × output price. Prices per 1M tokens as of October 18, 2026.

With 20 such tasks a month, multiply: about $0.07 on deepseek-v4.1-flash, about $0.45 on qwen3.8-max, $2.40 on claude-sonnet-5-5. This is a teaching example, not a forecast: a real agent hands the model the whole accumulated conversation at every step, so input will be noticeably larger. Caching does not reduce this equally everywhere: for GPT models the gateway sees cache reads in both formats and bills them at their own table price, while for DeepSeek, GLM and Qwen a cache read in Zukko costs the same as input. For Claude the cache discount works in the Anthropic format; more in the article on prompt caching↗.

To keep spending under control, each key in the dashboard can have a daily and monthly dollar limit.

What to choose

• You need only Qwen and are ready to go to Alibaba for a predictable bill: check their plans on their site.
• You need a choice of model families and one balance: a custom provider, for example a gateway.
• You just want to try: top up the minimum $1 and take a cheap model from the table. Judge quality on your own task.

We do not keep lists of "free keys" and advise against looking for them: someone else's key can stop working at any moment, and its owner will see your requests.

FAQ

Did Qwen Code become paid?

The program itself remains open. What became paid is model access: the free Qwen OAuth login, per the project documentation, was discontinued on April 15, 2026.

Can I run Qwen Code with no spending at all?

The only option with no payment to a provider is a local model on your own computer. Then you pay with hardware and electricity, and quality and speed depend on the model and the machine. Qwen Code can work with local models (the README mentions Ollama and vLLM). Pros and cons of this approach: local model or API↗.

Which model in Qwen Code is the cheapest?

In the Zukko catalog as of October 18, 2026 the lowest price is deepseek-v4.1-flash: $0.013 per 1M input and $0.038 per 1M output. That does not mean it will suit your task; test it.

Do I need a Zukko subscription?

No. The balance is shared and charges are per token.