Introducing Hy4 Preview

Tencent has released Hy4 Preview, a large open-weight text input language model with 770 billion total parameters, 49 billion active parameters, and a one million token context window. The model file size is 1.56 terabytes on Hugging Face, marking a substantial increase from the company's previous Hy3 model released in July, which had 295 billion total parameters, 21 billion active parameters, and a 256,000 token context window.

An examination of the model chat template on Hugging Face reveals two configurable reasoning effort levels named high, which serves as the default, and no_think, which disables reasoning. Testing the model with a prompt to generate an SVG of a pelican riding a bicycle using the default high reasoning setting on OpenRouter exposed a hidden reasoning trace.

The output showed the model internally debating design choices, such as adding a helmet or sunglasses, using slightly truncated English rather than perfect grammar. This release provides insight into the scaling and internal reasoning processes of large language models developed by Chinese technology companies.