Alibaba releases Qwen 3.8 27B vision model under open-source licence
Simon Willison’s initial assessment of the Apache 2.0-licensed model highlights strong performance alongside a tendency to overthink, as Alibaba continues to expand its large language model portfolio.
Alibaba’s Qwen research lab has released Qwen 3.8 27B, a 27-billion parameter large language model with vision capabilities. The release, which occurred on Friday, 16 August 2026, makes the model available under the Apache 2.0 open-source licence. This licensing framework permits broad commercial and non-commercial use, modification, and distribution of the technology.
The model integrates image processing capabilities with text-based language understanding, positioning it within the growing sector of vision-capable AI systems. By adopting the Apache 2.0 licence, Alibaba has enabled developers and institutions to utilise the model for a wide range of applications without restrictive usage limitations.
Early critical assessment of the model has been provided by reviewer Simon Willison, who published his findings on the same day as the release. Willison described the model’s performance as excellent, noting its potential for advanced tasks. However, he also highlighted a specific behavioural trait, observing that the model defaults to wildly overthinking things.
The characterisation of the model as excellent and the observation of its overthinking behaviour reflect Willison’s initial evaluation. These insights provide preliminary data on the model’s operational characteristics, though they represent a single reviewer’s perspective rather than a consensus view across the broader artificial intelligence community.
Claims regarding the model’s performance should be treated as preliminary until corroborated by independent technical evaluations. As Alibaba continues its ongoing development of large language models, the release of Qwen 3.8 27B adds to the competitive landscape of open-source AI tools available to researchers and developers.
