Qwen Code

You will like that it follows you: one agent in the terminal and inside VS Code, JetBrains and Zed, with /model to switch backends mid-session, so the tool you learned at the shell is the tool in the editor. The setup is the price: nothing happens until you pick a provider in the /auth menu, and that menu is where first-day enthusiasm goes to wait.

Pick it if you already hold keys and work across editors. Pick Claude Code if you want the best model with zero setup, and OpenCode if you want the same model freedom with a larger crowd behind it.

The risk is a resident process without isolation. Daemon mode keeps the agent running between sessions, there is no container sandbox in the product, and every shell command lands in the checkout you are editing; a resident agent with a shell has your permissions and does not log out when you do.

Run it inside a container of your own or not at all in a repository you cannot rebuild. What it does right: configuration precedence is documented in order, CLI flags, then environment, then .env files, then settings.json, so when a setting wins you can say why.

The lineage is explicit: Qwen Code is a fork of Gemini CLI, so the loop, context gathering and edit application are inherited rather than designed here. Documented additions: 1. SubAgents with their own context, which bounds what one task can pollute. 2. Agent Teams for orchestration across them. 3. Auto-memory and skills that persist across sessions, so context accrues rather than restarting.

No benchmark is published for the fork. The observation: a fork inherits the parent's bugs at no extra cost, and its fixes only when someone merges them.

Alibaba Cloud is the owner, which makes the tool a funnel for Qwen model spend rather than a business on its own. The tell arrived on 2026-04-15 when the Qwen OAuth free tier was discontinued: subsidized tokens ended, paid plans began, the subsidy-to-subscription turn inside a year. Twenty-seven thousand stars is real distribution, bought with those tokens.

No acquirer is needed; the parent is the acquirer. The pivot to watch is the model list narrowing toward Alibaba's own. Position: use it, expect the funnel to tighten, keep your keys portable.

The demo is a terminal agent. Procurement: no seat price, no contract, no SSO, no audit log, and model access through Alibaba's Coding Plan means choosing the Beijing or the international endpoint, which is a data-residency question, not a config detail. Sixty laptops pointing at the wrong one is a finding.

Headless mode gives CI a path, and the SDKs mean a platform team could wrap it, but nothing central records what ran. Onboarding is an afternoon. Security signs off on the endpoint first, and legal reads the Coding Plan terms. Not yet.

This is ownership. Apache-2.0 source, OPENAI_BASE_URL and OPENAI_API_KEY to point it at anything OpenAI-compatible including a local server, ANTHROPIC_API_KEY and GEMINI_API_KEY for the others, and a modelProviders block in settings.json to name them. MCP servers take a command or an httpUrl with headers, in the same file.

A fork survives: the license makes that a decision, not a negotiation, and the source builds without a vendor account. Alibaba could vanish tomorrow and my fork would not notice, which is the only longevity I trust.