← Back to stories Close-up of an MRI scan showing a sagittal view of the human brain for analysis.
Photo by MART PRODUCTION on Pexels
凤凰科技 2026-03-18

Tencent's HY 3.0 (混元3.0) Enters Internal Tests, Planned Public Launch in April 2026

What Tencent says

It has been reported that Tencent (腾讯) is testing the next major iteration of its in-house large language model, HY 3.0 (混元3.0), and plans a public launch in April 2026. Tencent describes this as a significant version jump from HY 2.0 (混元2.0), with marked improvements in model inference speed and in-built agent capabilities. The company says the upgrade represents an overall lift in the model’s “intelligence level” — a concise claim, but one that matters in a crowded Chinese AI field.

How the upgrade was built

Reportedly, Tencent’s mixed-yuan (混元) team restructured its organization and R&D processes starting in the second half of 2025 to accelerate iteration. The firm says it focused on raising training-data quality and rebuilding pre‑training and reinforcement‑learning infrastructure to support faster cycles and a more robust base model. Faster iteration, better data and stronger tooling: that’s the technical recipe Tencent is pointing to. But can engineering changes alone translate into real-world gains against leading Western and domestic rivals?

Leadership and stakes

Public records show Yao Shunyu (姚顺雨) joined Tencent in September 2025 and has since been named the company’s chief AI scientist, reportedly overseeing the HY series development. HY 3.0 is expected to be the first major product release led by Yao — a high‑visibility test of his technical leadership and Tencent’s ability to ship flagship AI capabilities under new management.

Why it matters beyond China

Why should Western readers care? China’s big tech firms — including Baidu (百度), Alibaba (阿里巴巴) and Huawei (华为) — are racing to field competitive base models for search, cloud and consumer products. At the same time, geopolitical pressures such as U.S. export controls on advanced chips and restrictions on AI collaboration shape how quickly Chinese firms can scale compute and training. HY 3.0 will be read not just as a product milestone, but as a signal of how China’s AI stack is evolving under both domestic competition and global trade constraints.

AISpace
View original source →