Qwen Code Desktop v0.3.0-preview.0 landed on 10 September with a notably connective upgrade: it can generate content through the OpenAI Responses API and delegate a subagent turn to an external agent over ACP, with Claude Code first in line. For developers, Qwen’s desktop workbench is widening beyond a single-agent, single-provider affair.
Alibaba Qwen Watch analysis
What happened
The preview also adds unified session sources, puts newly discovered models first, and improves several awkward edges of agent work. The changelog lists protection for pre-aborted queued tool requests, clearer matching deny rules for compound shell denials, duplicate cold-session restoration fixes and preservation of reasoning signatures during history consolidation.
Our top picks
- Responses API connection
Qwen Code can now generate content through OpenAI’s Responses API, adding another route for model-backed coding workflows. - ACP delegation
A subagent turn can be handed to an external agent, with Claude Code named as the first supported target. - Unified session sources
Sessions from different sources are brought into a more coherent view, making agent history easier to navigate. - Safer queued tools
Pre-aborted queued tool requests are rejected instead of lingering in the execution queue. - Clearer shell denials
Permission failures now cite the matching deny rule for compound and virtual-operation shell requests. - Fewer restoration oddities
The release fixes duplicate cold-session restoration and retries completed response delivery. - Desktop packaging progress
The release notes say macOS and Linux desktop packaging is unblocked.
Why it matters
The shift is not one shiny interface feature. Qwen is tightening the connective tissue around an AI coding agent: providers, sessions, permissions, external agents and desktop delivery all get attention in the same preview.
That matters in real repositories, where a working hand-off, a tool that fails closed and a session that comes back once are worth more than a dramatic demo. It is still a preview, so treat it as a test candidate rather than a universal upgrade command.
Our read
This is a meaningful workbench release for Qwen Code users who want interoperability and better operational edges. Try it against a disposable project first, then test the features you actually depend on: Responses API integration, ACP delegation, session restoration and shell permissions.
What to watch
- Whether ACP delegation expands beyond Claude Code and how approval boundaries behave across the hand-off.
- Which models and account configurations the Responses API generator supports in practice.
- Whether the desktop packaging fixes translate into stable signed builds across macOS and Linux.
- Any regressions in session restoration or queued-tool handling during longer agent runs.
Discussion spark: Which matters more for an AI coding workbench: handing tasks to other agents, or making sessions and permissions reliably boring?
Sources and evidence
- Qwen Code Desktop v0.3.0-preview.0 (10 September 2026, 05:53 UTC)
not affiliated with or endorsed by Qwen