Rumors say Qwen4 launches next week after Qwen's CTO remarks
Reports say Qwen's CTO pointed to a launch next week and a code change for the Qwen4 family has landed, but neither is confirmed.
Reports that Alibaba's Qwen team will launch Qwen4 next week rest on a secondhand quote and a conference slide. Neither amounts to an announcement. The commentator kimmonismus wrote that Qwen's CTO reportedly said Qwen4 would arrive "next week" and that a code change for the Qwen4 family has landed. The most concrete material is the slide. A Reddit post shared a screenshot of a slide shown at an Apsara event that reads "Qwen4 Series Coming Soon" and lists four variants: Max, Flash, Plus and 27B. The slide establishes names and a rough lineup. It doesn't give parameter counts, context length, architecture, benchmarks or pricing, and it doesn't give a release date. "Coming soon" describes a plan, not a schedule. The other thread is a code change. A Reddit post pointed to pull request #15762 in the qwen-code repository as evidence that the Qwen4 models will support a 1M-token context window. The same post said the pull request didn't mention Qwen4-27B. Its reading of a Max release date moved from November 1 to what it judged a likely October 1 or a placeholder. A pull request shows what someone committed to a repository. It doesn't show what the company will ship or when, so a context-length figure taken from it is an inference. Commenters drew their own conclusions from the variant names. Several expected Max and Plus to be available only through an API, and expected Flash to need high-memory systems, roughly 128GB or more, to run locally. The 27B was the variant most people thought could matter for local use. One user noted that a 27B at Q4 quantization is still slightly too large for a 32GB RAM Mac, and pointed to smaller mixture-of-experts models such as Qwen 3.6 A3B as more practical there. These are expectations based on naming and hardware, not statements from Alibaba. Licensing is another open question. Some commenters hoped the new models would keep the permissive MIT or Apache-style terms of earlier Qwen releases. Others cited Qwen 3.8 Flash Next as having a license that, in their reading, didn't say users own their code and data outputs. The Qwen4 materials don't address license terms.
Why it matters
If a 27B variant ships, it could matter for teams that run models locally, but nothing official has been published yet.
What you can do
Hold off on planning evaluations or procurement until an official release and license are published.
Original source
We have no link to the original source for this story yet.
Where we saw it
Latent Space
AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works