Hy4
Tencent's open-weight flagship model from August 2026, with 770 billion total parameters and a context window above one million tokens.
Hy4 is the fourth generation of Tencent’s Hunyuan model family, released as a preview in August 2026 with its weights published openly. It has 770 billion parameters in total but only around 49 billion active for any given request, using a mixture of experts design that keeps the running cost far below what the headline number suggests. Its context window, the amount of text it can hold in mind at once, is above one million tokens.
Tencent aimed it at everyday professional work rather than benchmark records: long coding sessions, spreadsheets and documents, financial analysis and research. The company also says the model took part in optimising its own training and inference code, which produced a 31.8 percent throughput gain. That claim is real engineering, but it describes a model suggesting improvements that humans then evaluated, not a system rebuilding itself.