Tencent Hunyuan released the open-source Hy4 preview on Aug. 28 on Hugging Face, with 770 billion total parameters, about 49 billion active parameters and a 1-million-token context window. The model more than doubles Hy3's 295 billion total parameters and roughly doubles its active parameters, with Tencent citing gains from pre-training, post-training, data scale and infrastructure across 12 benchmarks. Hy4 preview scored 85.4 on Terminal Bench 2.1, 64.3 on DeepSWE, 74.1 on Toolathlon-Verified and 37.1 on APEX-Agents, while an internal blind test placed it slightly ahead of GLM 5.3 and Kimi K3. It targets software engineering, office and financial analysis, game development and scientific research, and is rolling into products including CodeBuddy and WorkBuddy, with API pricing at 6 yuan per million input tokens and 18 yuan per million output tokens. Tencent said the early release can overthink complex questions and over-verify answers and lacks vision, and is meant to gather feedback before the final Hy4 launch as the company, led in AI by former OpenAI researcher Yao Shunyu, steps up investment.