Jin10 reported on June 30 that Tencent has open sourced the latest version of the Hunyuan model, Hunyuan-A13B. Hunyuan-A13B is an expert mix model with 80 billion parameters, of which 13 billion parameters are active. It supports both fast and slow thinking modes. Notably, Hunyuan-A13B has been specifically enhanced for AI Agents, creating an "adaptive brain" to respond to different scene changes, designing over 30 types of agent instructions, and combining variations in tools, actions, and responses to create 20,000 format combinations. According to the test data published by Tencent, Hunyuan-A13B achieved high scores of 87.3 and 76.8 in the mathematics AIME2024 and AIME2025 tests, surpassing DeepSeek-R1's scores of 79.8 and 70, and also significantly exceeding OpenAI's o1 model.
The content is for reference only, not a solicitation or offer. No investment, tax, or legal advice provided. See Disclaimer for more risks disclosure.
Tencent Open Source New Version of Hunyuan Model
Jin10 reported on June 30 that Tencent has open sourced the latest version of the Hunyuan model, Hunyuan-A13B. Hunyuan-A13B is an expert mix model with 80 billion parameters, of which 13 billion parameters are active. It supports both fast and slow thinking modes. Notably, Hunyuan-A13B has been specifically enhanced for AI Agents, creating an "adaptive brain" to respond to different scene changes, designing over 30 types of agent instructions, and combining variations in tools, actions, and responses to create 20,000 format combinations. According to the test data published by Tencent, Hunyuan-A13B achieved high scores of 87.3 and 76.8 in the mathematics AIME2024 and AIME2025 tests, surpassing DeepSeek-R1's scores of 79.8 and 70, and also significantly exceeding OpenAI's o1 model.