According to the official announcement from Tencent Hunyuan, on December 5th, Tencent officially unveiled the latest iterations of its Hunyuan Large Model: Tencent HY 2.0 Think and Tencent HY 2.0 Instruct. These new models are built on a Mixture of Experts (MoE) architecture, featuring a staggering total of 406 billion parameters and 32 billion active parameters, while supporting a context window of up to 256K. Their reasoning prowess and efficiency stand at the pinnacle of China's AI landscape.

Specifically, HY 2.0 Think has refined its pre-training data and reinforcement learning strategies, enabling it to excel in deep contemplation and the handling of intricate tasks. On the other hand, HY 2.0 Instruct is tailored for swift responses, making it ideal for scenarios demanding concise feedback. Presently, these models are accessible on the Tencent Hunyuan official website, Yuanbao, Ima, and other applications, with API interfaces made available through Tencent Cloud.