Full and lightweight versions previewed
China's Tencent announced on Oct. 29 that it will soon release the full version of "Hy4," the next flagship of its "Hunyuan" AI models, along with a lightweight version, "Hy4 flash." Senior Executive Vice President Dowson Tong (汤道生) disclosed the plans at the company's event, the Tencent Global Digital Ecosystem Summit (腾讯全球数字生态大会). @LeroyLi311063, an X account that covers AI developments in China, was among the first to report it.
Compared with the "Hy4 preview" released in August, the full version improves the ability to handle long-running real-world tasks and the efficiency of its reasoning. Hy4 flash prioritizes low cost while retaining image understanding and agent capabilities. It is the first in the Hy4 series to support image understanding. Tencent did not give a release date or pricing.
A large model with 770 billion parameters
The Hy4 preview has 770 billion total parameters, of which only 49 billion are used for each token. It can take in 1 million tokens at once.
On Nov. 2, @aitrackerbot, an account that tracks new models, reported that steps for running the full version of Hy4, not the preview, had been added to a proposed change to "vLLM-Ascend," inference software for chips from China's Huawei. The model reportedly accepts image input and has a 10-billion-parameter layer that predicts upcoming tokens together. The weights have not yet been released.
Official account posts a teaser
Tencent's official Hunyuan account on X posted a teaser for Hy4 on Nov. 3 (the evening of the same day, Japan time). The only accompanying words were "4 Real Productivity," and no release date was given. Expectations are growing that the full version and Hy4 flash will be released soon.
