Alibaba Publishes Qwen3.8-Flash-Next as an Early Look at Qwen4

Alibaba's Qwen team published Qwen3.8-Flash-Next on August 26, 2026, describing it as the first open-weight release of the architecture that will underpin Qwen4.

The model card lists 125B total parameters with 6B activated per token, alongside a 51B n-gram embedding table and a 4B multi-token prediction module. It runs 48 layers with a hybrid attention stack combining Gated DeltaNet and Qwen Sparse Attention, and a mixture-of-experts layer holding 512 experts with 11 active per token. Context is 262,144 tokens natively and extensible to 1,000,000. Input is text, image and video, through a vision encoder.

The licence is qwen-community-1.0. Qwen3.8-Max had its weights published earlier in August. "Open weights" and "Apache 2.0" are separate claims, and only one of them applies here.

Reporting on the launch, including TechNode, puts the commercial Qwen3.8-Flash API build at $0.16 per 1M input tokens and $0.47 per 1M output. Qwen has not published those rates on a pricing page we can point at, so treat them as reported rather than confirmed.