Beating AI News Flash: Tencent's Chief AI Scientist Yao Shunyu and the Hunyuan research team held a public Q&A on Reddit, revealing the R&D plan for the next-generation Hunyuan model.
A developer asked why Hunyuan remains at a scale of approximately 700 billion parameters, while vendors such as ByteDance and Kimi are exploring large models with trillions of parameters.
Tencent responded that a smaller but solidly trained model can surpass a larger model, and training costs are lower. However, the team still believes in the Scaling Law; the next-generation Hy model will significantly expand parameter scale, and the existing Hy4 will also continue to improve its capabilities through large-scale reinforcement learning.
Tencent also revealed a roadmap for small models that can run locally on personal computers. Hy4-mini is already in the plan, and the lighter Flash-Lite will also be launched.
