AI
Alibaba ships Qwen3.8-27B under Apache 2.0
Alibaba published the weights of Qwen3.8-27B on Hugging Face and ModelScope under the Apache 2.0 licence, with a 262,144-token native context that can be extended to one million tokens. The release is the deployable half of the Qwen3.8 family: a 27-billion-parameter dense multimodal model that Alibaba says outperforms its previous open releases on coding and computer-use benchmarks, while a larger 2.4-trillion-parameter mixture-of-experts sibling ships under a separate commercial licence.