Alibaba's Qwen team has released open weights for their Qwen3.8 models, fulfilling a promise to the community. The new Qwen3.8-27B is a native multimodal dense model with 27 billion parameters. It surpasses the previous Qwen3.7-Plus in overall performance and excels in practical coding and office tasks.

This model boasts a native context length of 262K tokens, which can be extended to 1 million tokens using YaRN. Designed for developers, Qwen3.8-27B is highlighted for its efficiency, quality, and the permissive Apache 2.0 license.

Alongside the 27B version, the open weights for the larger Qwen3.8-2.4T-A95B (Max-level) model have also been made available. Users can now download and deploy these models for various applications, from lightweight local deployments to building sophisticated AI agents.