Tencent released a preview of a new open-source AI model on Friday, Reuters reported, citing a post on Hugging Face. The model, Hy4 preview, is aimed at software engineering, research, and financial analysis.
Hy4 preview uses a mixture-of-experts design with 770 billion parameters in total, Tencent said, according to Reuters. About 49 billion of those parameters are used for any given text request. That is a large sparse model, not a 770 billion-parameter dense one. Tencent listed the stock as 0700.HK in the Reuters account.
The company said it plans to put the model into Tencent products, including CodeBuddy and WorkBuddy. Reuters did not describe those as already running Hy4 for every user. Integration is the plan. The weights are the Friday news: a preview posted to Hugging Face, the public repository for open models.
Tencent also flagged the usual preview caveats. The early-release model can take longer than it needs to work through complex questions, Reuters reported, and it may over-verify its own answers. That is Tencent saying the preview can stall and second-guess itself. It is not a finished production claim.
The release lands as Tencent spends more on AI in a crowded field, Reuters said. In April the company unveiled Hunyuan 3.0, its first major model release after hiring former OpenAI researcher Yao Shunyu to lead AI platform development.
Open-source Chinese models have become a regular Friday posting. Hy4 preview is Tencent’s latest entry in that stack, sized for coding and analysis jobs rather than a chat toy. The headline numbers are the 770 billion total and the 49 billion active. The useful warning is the one Tencent included: it may spend too long checking a correct answer.
A preview that over-verifies is still a preview. The weights are up. The products come next.
Sources