# Tencent Hunyuan Launches New Flagship Model Hy4 Preview

- Category: Flash (快讯)
- Published: 2026-08-28T12:00:00.000Z
- Source date: 2026-08-28
- HTML: https://www.88z.ai/en/news/flash-20260828-64bcd45e13e1f8c2
- Markdown: https://www.88z.ai/en/news/flash-20260828-64bcd45e13e1f8c2.md
- Other language: https://www.88z.ai/news/flash-20260828-64bcd45e13e1f8c2.md

Tencent Hunyuan has released its new flagship model, Hy4 preview, featuring 770B total parameters and 49B activated parameters with a 1M context window. The model excels in coding, office tasks, and scientific research, boosting inference throughput by 31.8% and achieving breakthroughs in molecular dynamics simulation speed and geometric problem-solving.

## At a glance

- Hy4 preview features 770B total parameters and a 1M context window, with a 31.8% boost in inference throughput.
- The model doubles molecular dynamics simulation speed and advances the 3D Blaschke-Lebesgue volume lower bound to 0.41104.

## The update

Tencent Hunyuan has officially launched its new flagship model, Hy4 preview. This model features a total of 770B parameters with 49B activated parameters and supports a context length of up to 1M. In real-world productivity applications, Hy4 preview demonstrates exceptional performance in tasks such as coding, office work, game development, and scientific research, achieving a 31.8% increase in end-to-end inference throughput. In the field of science, the model has achieved significant breakthroughs. It has doubled the speed of molecular dynamics simulations and made progress in solving a century-old geometric problem by advancing the lower bound of the three-dimensional Blaschke-Lebesgue volume to 0.41104. These achievements highlight the strong capabilities of Hy4 preview in handling complex computations and high-precision scientific tasks.

## Sources

- [腾讯混元](https://mp.weixin.qq.com/s/ymr3X878B8oa2XP15CH8TQ) (2026-08-28)

This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.