On September 22, the Tencent Hunyuan team announced the launch of the Hy Image3.5 preview generative-image model, positioned for professional-grade generative-image creation and supporting text-to-image, image-to-image, and multi-turn conversational creation. Unlike the common path of "release the official version first, then iterate," Tencent claims the model has already been gradually rolled out in gray-scale and internally tested in the Yuanbao App about 2 months ahead of time, with overall generative-image request volume up more than 50%. This means the preview is not a pure technical preview but is already handling real user traffic.
Tencent Hunyuan also provided a cadence explanation: since rebuilding its infrastructure in February, the Hunyuan large model has maintained a rhythm of iterating a major version on average every two months, and has adopted a preview-first, official-version-follows approach to bring real feedback into the R&D process. This statement explains why a preview version would enter the product side before the official release.
What needs to be distinguished is this: the above timing, gray-scale duration, and request-volume growth are all disclosed unilaterally by Tencent, sourced from release information reposted on a WeChat public account, and belong to source_claim, with no independent data corroboration yet.