Announced 9 Oct 2026 · Sources checked
What did the 9 October weekly actually ship?
The dated prose object is the Qwen Code docs article titled “Qwen Code Weekly: Qwen Live Is Live, Work the Pages You're Already Logged Into, Assign Tasks by Email,” attributed to Qwen Team and dated 2026-10-09. The footer we opened says “Last updated on October 10, 2026.” The lede says Qwen Code shipped four stable releases over the past two weeks—v0.24.5 through v0.25.0—with 467 PRs merged.
Those version strings line up with the npm registry we queried: @qwen-code/qwen-code 0.24.5 is timestamped 2026-09-24T17:16:56.822Z and 0.25.0 is timestamped 2026-10-05T09:55:55.980Z, with dist-tag latest pointing at 0.25.0. GitHub’s releases API for QwenLM/qwen-code also lists 10 October nightlies and v0.25.1-preview.2 published 2026-10-10T18:14:33Z. The weekly is the feature narrative; the registry and release tags are the installable objects.
For another October coding-agent stack that shares a CLI with people, see Alibaba Cloud’s ANOLISA v1.0 note, which cites Qwen Code sessions in a vendor token claim. For a different vendor’s local agent containment path, see GitHub Copilot local sandboxing GA.
What is Qwen Live, and what does it need?
The post says Qwen Live is available in both the Web Shell and the desktop version. Setup is explicit: open the Live card from the left sidebar, paste a Bailian DashScope Realtime API key, save, then start Open Live Voice from the Live page or the composer’s real-time voice control. During a call you can turn on screen sharing so the model sees what is on your display. The write-up points at pull requests #12529, #12189, #12378, and #12623 for those pieces.
Stated uses are hands-busy next steps spoken while typing, asking about an on-screen error after sharing the screen, and finishing a long back-and-forth by talking. Those are product scenarios in the weekly, not timings we measured. The Realtime API key requirement matters for anyone who expected Live to work on a default Qwen login alone; the post we opened does not claim a keyless Live path.
A related fix table row says that starting a new task during a Live call used to try to open another voice session and fail with “Live Voice is unavailable”; it now creates a normal task. That is a changelog claim tied to PRs #12516, #12519, and #12626, not a bug we reproduced.
How does Qwen Code use the Chrome you already signed into?
The weekly’s second headliner is browser control inside the Chrome profile you already use. After installing the Qwen Code Chrome extension from the Chrome Web Store, you can ask the agent to read open tabs, fill forms, and submit while staying signed into those sites. The post says this stays in the background with separate tab groups per concurrent session.
Platform scope is narrow in the text we opened: “Works on Mac and Linux; Windows isn’t supported yet,” with PR #12510 cited. That is a hard availability limit for Windows-first teams. The post does not publish a success-rate table for form filling or a security audit of the extension’s permissions. Treat browser automation as a capability announcement with an OS matrix, not as a verified RPA replacement.
If you are comparing agent UI patterns rather than browser control, CopilotKit’s Open Intelligent UI drop is a different stack—agent-built interfaces rather than driving an existing Chrome session.
What do Email Channel and /batch-api change?
Email Channel is described as a way to assign work when you are away from the machine running Qwen Code: configure a mailbox, allowlist your address, then send or forward mail; the agent acts and replies, and a reply thread keeps context. The post lists commute tasks, forwarded bug reports with log attachments, and alert mail as scenarios, citing PR #12939. Allowlisting is the access control the write-up names; we did not review the mail parser.
`/batch-api` targets piles of independent jobs—translating a docs directory, extracting fields from many tickets, tagging feedback. You describe the job; Qwen Code samples items and returns a free plan (counts, output path, rough tokens); work starts only after approval; billing is “at half the real-time rate”; results land as files while the conversation stays usable. The example command in the post translates Markdown under docs/zh into docs/en. Multi-round discussion jobs are told to stay in a normal conversation.
Those half-price and free-plan lines are Qwen’s commercial framing in the weekly. We did not submit a batch or compare invoices. Keep Email Channel’s allowlist and batch approval gates in the same column as Live’s API key: concrete setup steps, not measured reliability.
What else should operators notice before upgrading?
The Trajectory tab splits a turn across Model requests, Tool calls, and Subagent requests, with zoom, collapse, and a “Real time” view that separates elapsed wait from running time. Channel work can be split by message prefix into separate sessions, with private and group access as distinct settings, and long group threads can roll over when a limit is hit—the post cites one DingTalk thread of 8,577 messages as the failure mode that motivated rollover.
Experience claims include cutting time-to-prompt roughly in half from about 2.5 seconds on a clean machine and cutting memory about 60% from 575 MB, plus a full Chinese settings panel. `/commit` drafts a commit message and stages relevant files while refusing secrets such as `.env`, but leaves push to the user. Those performance and UX numbers are Qwen’s; we did not profile a cold start.
Upgrade instruction printed at the end of the weekly is `npm i @qwen-code/qwen-code@latest -g`. Pair that with the npm timestamps above if you need to know which published latest you are pinning. Preview and nightly tags on GitHub are separate channels from the stable weekly narrative.
| Claim | Where stated | What that is not |
|---|---|---|
| 467 PRs across v0.24.5–v0.25.0 | Weekly lede | A review of every merged PR |
| ~2.5s startup halved; ~60% less memory from 575 MB | Experience Improvements | Your machine’s cold-start profile |
| /batch-api at half real-time rate | /batch-api section | An invoice we reconciled |
| Chrome automation Mac/Linux only | Chrome section | Windows support |
| Live needs Bailian DashScope Realtime key | Qwen Live section | A keyless Live default |
Common questions
Is Qwen Live the same as the Qwen Code CLI?
No. The weekly presents Live as a voice-and-screen mode inside Web Shell and desktop Qwen Code, gated on a Bailian DashScope Realtime API key. The CLI/npm package is the broader product that also gained Chrome, email, and batch features in the same release window.
Can Windows users use the signed-in Chrome control today?
The post we opened says Mac and Linux are supported and Windows is not supported yet. That is Qwen’s matrix, not a test we ran on Windows.
Did Ai Lookout verify the half-price batch billing or the memory cut?
No. Those figures are stated in the weekly. Treat them as vendor claims until you measure your own sessions and bills.
What to remember
The 9 October Qwen Code weekly—last updated 10 October—bundles Live voice with screen sharing, signed-in Chrome automation on Mac/Linux, Email Channel, and half-price /batch-api into the v0.24.5–v0.25.0 line. Install from npm latest if that matrix fits; keep Qwen’s latency, memory, and pricing lines in the vendor column.
Sources & further reading
How this story was made
Written by Kristian Kostov with AI assistance and checked against the linked sources. Company performance claims are attributed to the company. Analysis reflects AiLookout’s interpretation; we have not independently tested the products discussed. Cover photography is illustrative and does not depict the specific announcement or product.
Our editorial standards
