Updated on
Qwen Code: reviews and analysis
Alibaba's terminal agent, now paid, with models that punch above their price.
Affiliate link · no extra cost to you
Our verdict
Qwen Code lost its free tier on 15 April 2026: the entry plan now starts at $50 a month internationally, or ¥200 in China. What remains is the performance: its code models still outperform what the rest of the category charges for the same work. On long reasoning and complex codebases the commercial leaders keep their distance, and the documentation trails in English and Spanish.
Best for: Developers wanting competitive code models without paying the commercial leaders' rates.
What changed this week
Sweep of
No substantive change from the previous sweep: what we checked this week confirms what was already there.
- no change
- no change
We have tracked it since July 25, 2026: 3 sweeps on record. We publish the latest change here; the full series is not published.
What the internet says
Qwen Code keeps shipping at a near-daily pace, with version 0.21.7 out on 6 August 2026 and the repo climbing from 26.3k to close to 26.8k stars in two weeks, while open issues rose from July's roughly 600 to 871. Access still has no free tier since it closed on 15 April 2026: the international paid plan runs from $50 a month and the Chinese one from ¥200, with the first month discounted to ¥39.90.
What the web repeats in favour
- The tool is open source under Apache-2.0 and the auth menu lets you pick the in-house provider, seven built-in third-party providers or your own endpoint
- Release pace stays high, with stable version 0.21.7 out on 6 August 2026, near-daily nightly builds and the repo climbing from 26.3k to close to 26.8k stars in two weeks
- It still supports models you run yourself, and the custom provider option connects any local endpoint that speaks the most common formats
- It has grown beyond the terminal: editor extensions, a continuous integration action, MCP, LSP, several development kits and messaging adapters
What the web repeats against
- The free tier of its own authentication was discontinued on 15 April 2026 and no longer even appears as an option in the menu
- The official paid door has harsh small print: the international plan runs from $50 a month with limited slots restocked daily and no refunds, and the middle Lite plan stopped selling on 20 March 2026 and stopped renewing on 13 April 2026
- Two official documents contradict each other: the tool's docs recommend the paid plan precisely for headless setups, and the provider's docs ban using that same key in non-interactive scenarios, with revocation as the penalty
- Price changes by region: the international tariff is $50 a month and the domestic one is ¥200 a month, with the first month at ¥39.90, different products rather than the same figure converted
- Open issue volume remains huge, at 871 as of today versus roughly 600 in July, and the release pace drags along unresolved interface bugs
Sweep sources: Official pricing · Docs · GitHub · GitHub · Docs · Official pricing
Pros / Cons
Pros
- Code models performing above their price despite the hike
- Seven built-in third-party providers, plus your own endpoint
- Open source, derived from Gemini CLI
Cons
- Lost its free tier on 15 April 2026: now paid from the first use
- Behind the leaders on long reasoning
- Less mature documentation outside Chinese
TLDR: Qwen Code competes with the simplest weapon: a free tier with thousands of daily requests on Alibaba’s code models, which perform far above what their price suggests. For bounded tasks and automation the math is unbeatable. On long reasoning the commercial leaders keep their distance, and documentation outside Chinese trails behind.
What Qwen Code is and how it works
Qwen Code is Alibaba’s entry into terminal agents, built as a derivative of Gemini CLI’s open source and adapted to the Qwen models, whose code family has long been surprising public leaderboards: performance near the frontier at bargain cost.
Its commercial argument is the category’s most aggressive free tier: thousands of daily requests without paying, an order of magnitude above what the competition gives away. For an individual developer, that limit is indistinguishable from unlimited.
As an agent it does the genre’s standard (tasks, project reading, editing, execution), with the maturity of the base it derives from and its models’ particularities.
What it is like day to day
For bounded work (specific features, fixes, scripts, test generation, batch automations), the experience surprises anyone arriving with prejudices: the Qwen models solve with solvency and the feeling is of a serious agent that happens not to charge.
The ceiling appears where it always does: long, ambiguous tasks where sustained judgment makes the difference, ground where Claude Code and company earn their subscriptions. The sensible usage pattern is the second agent: volume and automations to Qwen, the difficult to the leader.
The practical frictions: Western documentation and community still green, and the provider consideration every team should make explicitly according to its data policies, as with any service from a different jurisdiction than your own.
Pricing and plans
The free tier with sign-in gives thousands of daily requests, enough for complete individual use. Above it, pay-per-use on Alibaba’s API, with per-token prices among the market’s lowest. There is no subscription to manage.
The comparison with Gemini CLI, its free relative, gets decided by limits and models: Qwen gives away more requests, Gemini brings the long context and the Google ecosystem. Keeping both configured costs zero and is what many people do.
Who it is for (and who it is not for)
Qwen Code is for zero-budget developers with many bounded tasks, for volume automations where cost per task rules, and as the mechanical-work second agent alongside a paid leader.
It is not for anyone delegating long reasoning without watching cost, nor for teams whose data policies rule the provider out, a decision worth making consciously rather than by inertia.
Qwen Code alternatives
Gemini CLI is the free Western relative with long context. Kimi CLI competes in the same minimum-cost corner with a cheap membership. Claude Code and Codex are the paid judgment jump. They are in Qwen Code alternatives and the ranking in the best AI coding tools.
Frequently Asked Questions
Are the thousands of free daily requests real?
Yes, with sign-in: it is the category’s most aggressive free tier, and for an individual developer it works as unlimited. Pay-per-use only appears at industrial API volume.
How good are the Qwen models at code?
Far above what their price suggests: near the frontier on bounded tasks according to public leaderboards. The distance appears on long reasoning and autonomous exploration.
What is its relationship with Gemini CLI?
It derives from its open source, adapted to the Qwen models. They share base maturity and differ on models and limits: many developers keep both configured, because it costs zero.
Should I worry about the provider?
As always: read the data policy and decide for your case and jurisdiction. For personal projects and open source, most find no objection. For companies with requirements, it is the first question, not the last.