Claude Code, Qwen Code duel for local agents

If model quality is held constant, this is mostly a tooling fight, and Qwen Code has the cleaner story for zero-cloud setups.

• –Qwen Code explicitly documents local self-hosted model support through OpenAI-compatible endpoints like Ollama, vLLM, and LM Studio, with model switching handled in `settings.json` and `/model`
• –Claude Code can also run against local servers, but guides rely on environment overrides like `ANTHROPIC_BASE_URL` and extra tweaks such as disabling attribution headers to avoid local inference slowdowns
• –Qwen Code being open-source and optimized around Qwen’s own coding models makes it easier to tweak for hobbyist and self-hosted workflows
• –Claude Code still looks stronger on overall UX and agent workflow polish, especially if you already like Anthropic’s terminal ergonomics and IDE integrations
• –For users with no cloud plans at all, the thread’s strongest practical takeaway is that Qwen Code is usually the simpler default, while Claude Code is the better pick only if its workflow feels worth the extra plumbing