Basically Official: Qwen Image 2.0 Not Open-Sourcing
A website tag change confirms the powerful 7B DiT image model will not be publicly released.
Alibaba's Qwen AI team has signaled a major strategic pivot by reclassifying its Qwen Image 2.0 model from an open-source project to a closed-source product. The confirmation comes not from an official announcement, but from a subtle change on the Qwen research webpage. The blog post for the advanced image generation model, initially tagged 'Open-Source' upon its announcement in February, has now been updated to the 'Release' category. This tag is reserved for the team's commercial, closed-source offerings like the Qwen-Max series, effectively dashing hopes for a public release.
The Qwen Image 2.0 is a significant model, built on a 7B parameter Diffusion Transformer (DiT) architecture and paired with the Qwen 3 VL 8B text encoder. Its technical details had been shared, fostering anticipation within the AI community for its open-source release after the Chinese New Year. The reversal follows a period of uncertainty marked by team departures and aligns with a broader trend of AI companies opting to keep powerful models proprietary. This decision mirrors the earlier 'Wan 2.5 situation,' where another promising model was withheld, highlighting the growing tension between open research and commercial interests in the generative AI race.
- The Qwen Image 2.0 blog post tag changed from 'Open-Source' to 'Release,' a clear indicator of a closed-source strategy.
- The model is a 7B parameter DiT architecture using the Qwen 3 VL 8B text encoder, details of which were previously shared.
- This follows team departures and mirrors the earlier 'Wan 2.5' case, marking a trend away from open-sourcing cutting-edge AI.
Why It Matters
This move reduces accessible state-of-the-art image models for developers and researchers, concentrating advanced AI capabilities within large corporations.