What is Qwen Code?
Qwen Code is an open-source coding agent from Alibaba's Qwen team. It started as a fork of Google's Gemini CLI v0.8.2 and stopped syncing upstream at v0.1, and has since grown its own feature set: subagents and agent teams, plan mode, hooks, skills such as /review and /bugfix, MCP, language-server integration, git worktrees, a sandbox, and an Agent Arena that runs two models on the same task side by side. Besides the terminal it ships VS Code, Zed and JetBrains plugins, a desktop app, a background daemon, chat connectors for Telegram, DingTalk, WeChat and Feishu, and SDKs for TypeScript, Python and Java.
It speaks the OpenAI, Anthropic, Gemini and Qwen APIs, so it runs against any compatible provider or a local model through Ollama or vLLM. The free Qwen OAuth tier, which once gave 1,000 requests a day, was cut to 100 on 13 April 2026 and closed on 15 April. Model access now comes from Alibaba Cloud plans, another provider's key or local hardware. Releases are close to daily; v0.26.5 shipped on 1 November 2026.
Is Qwen Code worth it?
Qwen Code is now a broader agent than the Gemini CLI it was forked from, and it ships quickly. The repository has about 28,900 stars, 3,300 forks and 2,094 open issues and pull requests, and in our reading of the releases page it ships a tagged version most days: v0.26.4 on 29 October and v0.26.5 on 1 November, alongside desktop and SDK builds.
The project publishes an unusual consistency check. Seven versions from 0.0.14 to 0.22.0, each run three times on the 500 tasks of SWE-bench Verified with the same Qwen 3.7 Max model, scored between 76.40% and 77.80%. That says the harness has not regressed as features piled on, though it is the team's own measurement and says nothing about other models.
The free tier was the reason many people installed it, and it is gone. Alibaba cut the Qwen OAuth quota from 1,000 to 100 requests a day on 13 April 2026 and shut it on 15 April, citing cost. The supported route is now Alibaba Cloud's Coding Plan: Pro costs $50 a month for up to 6,000 requests per five hours, 45,000 a week and 90,000 a month, with Qwen 3.7 Plus, Kimi K2.5, GLM-5 and MiniMax-M2.5. Alibaba counts each model call, not each question, and says a complex task takes 10 to 30 calls. The terms forbid automated scripts and backends, so the plan cannot power CI, and slots are rationed daily.
Two defaults deserve attention. The daemon listens on 127.0.0.1:4170 with no bearer token on loopback, so any local process can drive the full API, code execution included; the README says to set QWEN_SERVER_TOKEN and --require-auth on shared machines. And `privacy.usageStatisticsEnabled` is true by default in the settings reference.
Against OpenCode or Crush, Qwen Code's case is breadth: chat-app channels, a Java SDK and Agent Arena are things neither offers. Against Claude Code or Codex, which it openly imitates, it trades a flat-rate first-party model for choice of provider.
Where it earns its place
A solo developer who used the old free tier
Keep using Qwen Code now that free requests have ended
Subscribe to Coding Plan Pro at $50 a month and point Qwen Code at its own `sk-sp-` key and endpoint; a regular pay-as-you-go key there bills separately. At 10 to 30 calls per complex task, 90,000 requests covers roughly 3,000 such tasks a month.
A platform team at a Chinese e-commerce company
Run an agent from DingTalk and Feishu threads
Start `qwen channel start` against a pay-as-you-go ModelStudio Token Plan, since the Coding Plan forbids backend use. Set QWEN_SERVER_TOKEN and --require-auth before exposing the daemon, which is open on loopback by default.
An engineering lead comparing models for a 15-person team
See which model handles our codebase best before standardising
Agent Arena runs two configured models on the same task side by side. The software is free; each run costs both models' tokens, which keeps the comparison cheap on open-weight models.
Before you commit
- Setup
- `npm install -g @qwen-code/qwen-code@latest` (Node.js 22 or later), Homebrew or the standalone install script; run `qwen` in a project and `/auth` to choose QwenCloud, Alibaba ModelStudio, a third-party provider or a custom endpoint.
- Integrations
- OpenAI, Anthropic, Gemini and Qwen-compatible APIs; DeepSeek, Kimi, MiniMax, Z.AI, OpenRouter and others as presets; Ollama and vLLM locally; MCP; LSP; VS Code, Zed and JetBrains; Telegram, DingTalk, WeChat and Feishu; TypeScript, Python and Java SDKs.
- Data and privacy
- Prompts go to the provider you choose. Usage statistics are on by default and can be turned off in settings; prompt logging in OpenTelemetry export is a separate setting. Apache 2.0.
- Cost at scale
- Coding Plan Pro is $50 per person a month and may not be shared, so a 15-person team pays $750 a month for interactive use; CI and automation must go through metered keys.
- Limits
- Coding Plan caps of 6,000 requests per five hours and 45,000 per week, daily-rationed slots, no refunds; the daemon is unauthenticated on loopback unless configured; 2,094 open issues and pull requests.
What’s good
- Apache 2.0, with releases most days
- Subagents, plan mode, hooks, skills, worktrees and MCP in one tool
- VS Code, Zed and JetBrains plugins, desktop app and TypeScript, Python and Java SDKs
- Own SWE-bench Verified runs show 76.40% to 77.80% across seven versions
- Works with OpenAI, Anthropic, Gemini and Qwen APIs or local models
What’s not
- Free Qwen OAuth tier closed on 15 April 2026
- Daemon has no authentication on loopback by default
- Usage statistics on by default
- Coding Plan bans non-interactive use, so CI needs pay-as-you-go keys
- 2,094 open issues and pull requests
Best for
Developers who want a feature-rich, Apache-licensed terminal agent and are happy to pay Alibaba or another provider for the model.
Qwen Code alternatives
Qwen Code: common questions
- Is Qwen Code worth it?
- Qwen Code has outgrown the Gemini CLI it was forked from, but without its free tier it is mainly a good front end for Alibaba's $50 Coding Plan, and its daemon ships with no password. We rate it 3.8 out of 5.
- How much does Qwen Code cost?
- Free and open source (Apache 2.0). Models cost extra: Alibaba Cloud Coding Plan Pro $50/month (up to 90,000 requests a month, interactive use only), pay-as-you-go Token Plans, any third-party API key, or local models. The free Qwen OAuth tier ended on 15 April 2026.
- What are the downsides of Qwen Code?
- Free Qwen OAuth tier closed on 15 April 2026; daemon has no authentication on loopback by default; usage statistics on by default; coding Plan bans non-interactive use, so CI needs pay-as-you-go keys; 2,094 open issues and pull requests.
- Who is Qwen Code best for?
- Developers who want a feature-rich, Apache-licensed terminal agent and are happy to pay Alibaba or another provider for the model.
- How much does Qwen Code cost for a team or in production?
- Coding Plan Pro is $50 per person a month and may not be shared, so a 15-person team pays $750 a month for interactive use; CI and automation must go through metered keys.
- Is Qwen Code safe for company data?
- Prompts go to the provider you choose. Usage statistics are on by default and can be turned off in settings; prompt logging in OpenTelemetry export is a separate setting. Apache 2.0.
- What are the best Qwen Code alternatives?
- The closest alternatives we have reviewed are OpenCode, Crush, Gemini CLI and OpenAI Codex CLI.
Answers checked Nov 2, 2026.