Alibaba's Qwen Code adds streaming JSON and multilingual UI

Most terminal AI tools are fun for casual chatting, but trying to pipe raw terminal text into automated scripts is usually a recipe for broken parsing.

Alibaba's Qwen Code team just rolled out a major update spanning v0.2.2 to v0.3.0, designed to make the tool far friendlier to automation. The headline addition is native Stream JSON support across both inputs and outputs. Backed by a new three-tier adapter architecture and dedicated session management, developers can now feed structured data into the tool and receive streaming JSON back, opening the door for clean SDK integrations and CI/CD pipelines.

Plumbing the terminal into automated workflows

Beyond structured streaming, the release introduces full internationalization. It includes built-in English and Chinese interfaces with runtime commands to switch the UI and AI response languages independently, alongside support for community language packs.

On the reliability front, the team added a 20MB buffer limit to prevent memory exhaustion crashes, resolved Windows character encoding corruption, improved ripgrep detection for cross-platform setups, overhauled authentication management, and added support for ModelScope providers.

Why it matters

Terminal coding assistants are everywhere, but headless tools that play nicely with production infrastructure are much rarer. Adding Stream JSON transforms Qwen Code from a human-driven CLI into a reliable building block for automated code reviews, linting, and deployment scripts.

We think the unglamorous plumbing work here is what actually counts. Fixing Windows encoding quirks and capping memory buffers won't dominate social feeds, but it makes the difference between an experimental open-source script and a utility developers can safely trust in unattended pipelines.