Tencent Unveils and Open-Sources Hy3 (Hunyuan 3.0) AI Model, Led by Former OpenAI Researcher

Tencent has officially unveiled Hy3 (Hunyuan 3.0), its most advanced AI model to date, marking a significant leap in complex reasoning and coding capabilities. The model boasts 295 billion parameters and employs a Mixture-of-Experts (MoE) architecture, which activates only relevant subsets of parameters per task to balance performance with efficiency. Under the leadership of Shunyu Yao, a former core researcher at OpenAI, Hy3 achieved an average score of 88.4 on Tsinghua University's 2026 spring mathematics Ph.D. qualification exam, ranking first among all Chinese AI models. This performance underscores its strength in handling advanced mathematical reasoning, a key benchmark for frontier AI systems.

In a bold move that contrasts with the increasingly proprietary stances of Western AI leaders, Tencent open-sourced Hy3 immediately upon announcement, making it available on platforms like Hugging Face, GitHub, and ModelScope. This strategy aims to accelerate adoption among developers and researchers, particularly in the Chinese tech ecosystem. While specific coding benchmarks were not detailed, Tencent emphasized that Hy3 excels in code generation and debugging tasks, positioning it as a direct competitor to models like OpenAI's GPT-4 and Meta's Llama 3. The release signals Tencent's ambition to lead in the open-source AI space, leveraging top talent and massive infrastructure.