Tencent has quietly released a powerful new AI model called Hy3-preview (also known as Hunyuan 3), designed to revolutionize autonomous coding and agentic workflows. Built by a team that includes former OpenAI researchers, Hy3 focuses on reasoning rather than sheer scale.
The model uses a 295B Mixture-of-Experts (MoE) architecture with multi-token prediction, and it has achieved an impressive 74.4% score on the SWE-bench (Verified) benchmark. A standout feature is its dual 'Reasoning Toggle'—allowing users to switch between low and high reasoning modes. In high reasoning mode, Hy3 can tackle complex software engineering tasks that challenge other large language models.
Early reactions have been mixed; one developer noted that while Hy3 is strong, another model 'big pickle' performed better for certain coding tasks. Nonetheless, Tencent's Hy3 marks a significant step forward in agentic AI, offering developers a new tool for automated coding and problem-solving.