AI BriefingTencent Releases 'Hy4 Preview', a 770B Parameter Open-Weight LLM
Key point
Tencent has released Hy4 Preview, an open-weight LLM with 770B parameters and 1M context support.
Details
Chinese tech giant Tencent has released Hy4 Preview, a new open-weight text-only LLM. The model adopts a MoE architecture with 49B active parameters out of 770B total, and supports a context window of 1M tokens. According to Hugging Face, the model file size is 1.56TB.
Performance and Specification Improvements Over Hy3
Hy4 Preview features significantly upgraded specifications compared to its predecessor Hy3, released in July. While Hy3 offered 295B total parameters, 21B active parameters, and a 256K context window, Hy4 increases the parameter count and context length by approximately 2.6x and 4x, respectively.
Reasoning Modes and Test Results
Analysis of the model's chat_template.jinja reveals that reasoning effort offers only two options: 'high' (default) and 'no_think' (reasoning disabled). Simon Willison conducted a test generating a "pelican riding a bicycle" via OpenRouter, observing that the reasoning process in the default 'high' mode was recorded concisely in English. This suggests a prioritization of token efficiency over perfect grammar in the hidden reasoning text.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Step 5 Preview Released: 600B Parameter MoE Model Highlights Agentic Tasks and Financial Analysis Performance