Lates News

On December 5th, Tencent's latest language models Tencent HY 2.0 Think and Tencent HY 2.0 Instruct were officially released. HY 2.0 adopts a Mixed Expert (MoE) architecture with a total of 406 billion parameters and 32 billion activation parameters, supporting a 256K context window for reasoning capabilities and efficiency that are among the best in the country, excelling in practical scenarios such as text composition and complex instruction following. Compared to the previous version (Hunyuan-T1-20250822), the HY 2.0 Think model has significantly improved in pre-training data and reinforcement learning strategies, stabilizing its position as one of the top performers in complex reasoning scenarios such as mathematics, science, code, and instruction following, with greatly enhanced generalization abilities. (Tencent Hunyuan).