Simon Willison • 8/29/2026

Introducing Hy4 Preview

This article introduces Tencent's Hy4, a new open-weight text-only LLM with 770B total parameters (49B active), a 1M token context window, and 1.56TB size on Hugging Face. It compares to the previous Hy3 model, highlighting a significant size increase. The author examines Hy4's chat template, revealing two reasoning effort levels: 'high' (default) and 'no_think' (disabled). They test the model with an SVG generation prompt, showing a reasoning trace that uses truncated English for efficiency. The post is a technical review of a new AI model, relevant to IT/technology.

0 comments

#llm #Reasoning #Openrouter