A report from the opening of the 2026 Yunqi Conference identifies four names in an upcoming Qwen4 family: Qwen4-Max, Qwen4-Flash, Qwen4-Plus and Qwen4-27B. The accompanying event photo labels the series “Coming Soon,” but does not provide a release date or technical specifications.

In a post on X, @MaxForAI says Liu Dayiheng, described in the post as Alibaba’s Qwen LLM lead, announced the upcoming Qwen4 family at the conference. The post includes a photograph of a presentation screen listing the model names.

The Qwen4 models shown in the report

The photograph groups the names into three entries:

  • Qwen4-Max

  • Qwen4-Flash and Qwen4-Plus

  • Qwen4-27B

The presentation identifies the names but does not describe how the models differ. It does not say whether “Max” or “Flash” refers to a particular capability, model size, speed target or deployment option. It also does not explain how Qwen4-27B relates to the other entries beyond showing it as part of the series.

The four names therefore represent the lineup visible in the report, not a complete product specification. The report does not establish whether they are final names or whether they represent every model planned for the Qwen4 family.

“Coming Soon” does not provide a launch date

The event slide says “Qwen4 Series Coming Soon.” That indicates an upcoming release in broad terms, but it does not identify a launch day, rollout schedule or expected availability window.

The report also does not say whether any of the models will be available through an API, downloadable for local use or offered only in particular regions. It provides no pricing, licensing terms, hardware requirements or context-length information. There are also no supplied capability descriptions, benchmark results or details about supported modalities.

For now, the report gives readers four Qwen4 names to track, but not enough information to decide which model would be suitable for a specific task or deployment.

What the separate “5-10T” reference means

The post separately says that Qwen will train “5-10T” models in the future. It does not define what “5-10T” refers to, so the phrase should not be treated as a confirmed Qwen4 specification or expanded into a claim about training tokens, model parameters or dataset size.

The report also does not establish whether the reference is connected to the Qwen4 lineup. More information would be needed before drawing that connection. Release timing, model documentation and technical details remain unknown.

Sources