Alibaba releases experimental Qwen Flash-Next model

Alibaba's Qwen team released Qwen3.8-Flash-Next, an open-weight experimental model available on Hugging Face under the qwen-community-1.0 license. The company describes the model as a preview of an architecture intended for Qwen4. It has 125 billion parameters but activates 6 billion per token, with a native 262,144-token context length that can extend to 1 million. Qwen says the design targets lower inference costs for long-context agent workloads; its benchmark results are vendor-reported and use its own evaluation choices.