Qwen3.8-27B under the Apache 2.0 licence
The Qwen team published a 27-billion-parameter multimodal model with a 262-thousand-token context under a permissive licence that allows commercial use without negotiation.
Why it matters
A model that understands images and video natively, with a quarter-million-token context, became freely available for commercial use.
The weights were uploaded to the Hugging Face repository on 13 August 2026; the repository itself was created on 5 August and the licence file added on 14 August. The licence is Apache 2.0. The card states 27 billion parameters and a context length of 262,144 tokens, extensible up to one million. The model has native support for image and video understanding, from STEM diagrams and documents to hour-scale videos, and is classified as image-text-to-text. The card gives no release date, so the record rests on the commit history. It is the smaller companion to Qwen3.8-Max, whose weights were published a few days earlier under a different, custom licence.