Alibaba’s Qwen team has released open-weight versions of its Qwen3.8 models under the Apache 2.0 license, according to The Decoder. The flagship Qwen3.8-27B is a 27-billion-parameter multimodal model that the team says outperforms the larger Qwen3.7-Plus on coding and office tasks, while supporting up to 262,000 tokens of native context and multimodal input spanning text, images, and video. The models are available on Hugging Face and ModelScope, with a hosted version offering a million-token context window planned for Qwen Cloud.
