Qwen Code v0.23.3 adds an OpenAI Responses API content generator and lets a subagent hand a turn to an external agent over ACP, with Claude Code named as the first hand-off target. The practical point is a coding agent that is easier to connect, inspect and recover when a real repository gets untidy.
Alibaba Qwen Watch analysis
What happened
Published on 10 September, the release expands Kimi, Qwen and DeepSeek reasoning presets, adds unified session sources and improves session navigation and details. Answered questions also remain visible outside processed summaries.
Qwen lists fixes for pre-aborted queued tool requests, pending permissions and questions surviving refreshed session loads, richer Responses stream error details, Windows PTY handling and restored VS Code conversation history. The release notes list no known breaking changes.
Our top picks
- Responses API generator
Qwen Code can now generate content through the OpenAI Responses API, widening its integration surface. - ACP hand-offs
A subagent turn can be delegated to an external agent over ACP, with Claude Code listed as the first target. - Reasoning presets
Kimi, Qwen and DeepSeek presets are expanded with more built-in configuration routes. - Session visibility
Unified sources, improved navigation and visible answered questions make agent activity easier to inspect. - Permission continuity
Pending permissions and questions survive refreshed session loads instead of vanishing into the refresh. - Failure detail
Responses stream errors retain more detail, making integration failures easier to diagnose.
Why it matters
The interesting change is the combination. Qwen Code is not only adding another endpoint; it is trying to make an agent workbench interoperate with other agents while preserving context, approvals and failure detail.
Developers can inspect the Responses API path, test ACP delegation against existing tools and pay attention to the session and permission fixes if their workflow spans refreshes or parallel work.
Our read
This is the sort of release that can improve a coding agent’s daily texture more than its launch headline. The capabilities are promising because they connect the plumbing to the work, but ACP interoperability still deserves testing in your own repository.
What to watch
- Whether ACP hand-offs preserve enough context and approval state.
- Which external agents become practical beyond the named Claude Code path.
- Whether richer Responses errors speed up integration debugging.
- How session visibility holds up with concurrent or refreshed work.
Discussion spark: Would ACP hand-offs simplify your coding-agent workflow, or add more context loss than they save?
Sources and evidence
- Release v0.23.3 (10 September 2026, 14:59 UTC)
not affiliated with or endorsed by Qwen