Alibaba’s Qwen team has released open weights for Qwen3.8 models under the Apache 2.0 license, according to The Decoder and a Techmeme item citing @alibaba_qwen. The release includes Qwen3.8-27B, which Alibaba describes as a new dense 27-billion-parameter multimodal model. The performance claims should be read as vendor claims for now. Techmeme’s summary says Alibaba claims Qwen3.8-27B beats Qwen3.7-Plus and excels in real-world coding. The Decoder similarly reports that Qwen says the 27-billion-parameter model outperforms the larger Qwen3.7-Plus in coding and office tasks. The Decoder reports that Qwen is also pitching improved agent behavior, saying the model can plan more independently and complete tasks more reliably. That framing matters because coding, office work, and agent-style execution are increasingly used as practical benchmarks for whether a model can do useful multi-step work, not just answer prompts. On context length, The Decoder reports that Qwen3.8-27B natively handles up to 262,000 tokens and can scale to one million tokens using the YaRN method. The model is also described as multimodal: it can process images and videos, including diagrams, documents, and multi-hour video. The release includes a flexible thinking mode that is enabled by default and can be toggled per query, according to The Decoder. The summaries do not provide independent benchmark results or deployment cost data. Beyond Qwen3.8-27B, The Decoder reports that Alibaba also released weights for Qwen3.8-2.4T-A95B, a much larger model described as built to operate at the Max level. Both models are available on Hugging Face and ModelScope, according to the report. A hosted version with one million tokens of context is expected to come to Qwen Cloud, Alibaba’s AI service, The Decoder reports. For now, the concrete change is the availability of Apache 2.0 open weights for the Qwen3.8 line, with Alibaba positioning the 27B model as a smaller model that can challenge a larger predecessor in coding and office-oriented workloads. Who benefits: Developers and AI teams that prefer open-weight models benefit from another Apache 2.0 option with long-context and multimodal capabilities. Who's exposed: Too early to tell from the provided reports. The provided reports do not establish cost, latency, or benchmark trade-offs.