Key takeaways
- Alibaba said at its September 22, 2026 Apsara Conference that Qwen 4 is in training.
- Alibaba has given no release date, size, price or benchmarks for Qwen 4.
- As of October 29, 2026, no Qwen 4 model is listed on Qwen's Hugging Face page.
Alibaba has not announced a Qwen 4 release date. At its Apsara Conference in Hangzhou on September 22, 2026, the company said Qwen 4, its next-generation model, is in training, according to Alibaba's own Apsara roundup. It gave no launch date, price, size or benchmark scores.
That matters because Apsara is already over. If you saw a claim that Qwen 4 will launch at the conference, the conference took place more than a month ago, and Alibaba's coverage of it describes a model still being trained.
What Alibaba has confirmed about Qwen 4
Alibaba's September 24 post says Qwen 4 is currently in training. It does not say how large the model is.
The same post says the later Qwen 4.5 and Qwen 5 series are projected to scale up to 5 to 10 trillion parameters. Parameters are the adjustable numbers inside a model, and more of them usually means a bigger and more expensive model to run. Alibaba gives no dates for Qwen 4, Qwen 4.5 or Qwen 5.
In a separate AliViews piece on the same conference, Alibaba says the Qwen team plans to train a new model at the scale of 5 to 10 trillion parameters. The goal it states is completing more complex, longer-horizon tasks. That piece does not name Qwen 4 and gives no training timeline.
What has shipped so far
Nothing called Qwen 4 is out. As of October 29, 2026, the Qwen page on Hugging Face lists no Qwen 4 model. The newest language models there belong to the Qwen3.8 family: Qwen3.8-Flash-Next, Qwen3.8-27B and Qwen3.8-2.4T-A95B, plus FP8 versions of each.
Alibaba's Apsara post also describes how the current flagship was used. It says Qwen3.8-Max completed 33 iterative self-improvement cycles over a month of automated runs, raising its Artificial Analysis score from 40 to 45. Those are Alibaba's figures, and we have not seen them checked independently.
How to read the roadmap
The two Alibaba posts together show a direction, not a schedule. Qwen 4 is the model being trained now, and the 5 to 10 trillion parameter scale is attached to later series. Alibaba says the team's stated aim is harder tasks that take longer to finish. Training a model is not the same as being ready to ship it, and Alibaba has not said how long testing and release will take after training ends.
What we don't know yet
- The release date for Qwen 4. Alibaba has given none.
- Whether Qwen 4 will be open-weight like the Qwen3.8 models on Hugging Face, or API-only.
- The model sizes, context window, API pricing and benchmark results.
- Any tier names. Alibaba's own materials do not list them, so lists of variant names circulating online are unconfirmed.
The place to watch is qwen.ai and the Qwen organization on Hugging Face, where Alibaba posts its releases. Until one of them lists Qwen 4, treat any date as a guess.
Frequently asked questions
When is Qwen 4 coming out?
Alibaba has not said. It confirmed on September 22, 2026 that Qwen 4 is in training and gave no release date.
Will Qwen 4 launch at the Apsara Conference?
The 2026 Apsara Conference was held on September 22. Alibaba described Qwen 4 as in training there and did not release it.
Is Qwen 4 available on Hugging Face?
Not as of October 29, 2026. The Qwen organization's newest language models are in the Qwen3.8 family.
How big will Qwen 4 be?
Alibaba has not said. It projects that the later Qwen 4.5 and Qwen 5 series will scale up to 5 to 10 trillion parameters.
