Tencent Mix Yuan: Hy3 preview launched for two weeks, Token calls increased tenfold

Zhitong
2026.05.07 05:47

Since its launch, the Token call volume of Tencent's Hy3 preview has grown significantly, reaching 10 times that of Hy2, especially in code and agent scenarios where it has increased by more than 16.5 times. The call volume of Hy3 preview on OpenRouter is 3.66 trillion, achieving the number one market share. This model adopts a mixture of experts architecture, with a total of 295 billion parameters, aimed at collecting user feedback to guide subsequent iterations

According to Zhitong Finance APP, on May 7th, Tencent Mixyuan announced that since its launch, the Token call volume of Hy3 preview has been continuously increasing, currently reaching ten times that of the previous version model Hy2. Notably, the Token call volume in code and agent scenarios has seen significant growth, with an increase of over 16.5 times in applications such as Tencent's WorkBuddy/Codebuddy and Qclaw.

In addition, public data from OpenRouter shows that Hy3 preview achieved a total call volume of 3.66 trillion Tokens in the past week, securing the top spot in both the weekly leaderboard and market share, and also ranking first in programming and tool calling scenarios.

The head of Tencent Mixyuan stated: "Hy3 preview focuses on practicality. To collect user feedback and understand the model's performance in real scenarios, Hy3 preview launched a limited free trial on OpenRouter at the beginning of its release, allowing developers to try the model for free. During this process, we have seen a continuous increase in the model's call volume, and we have indeed collected a lot of positive and negative feedback, which provides direction for the model's subsequent iterations. We would like to especially thank developers for their attention and recognition of Mixyuan."

It is understood that Hy3 preview is the first model launched after Tencent Mixyuan's technical restructuring, adopting a hybrid expert (MoE) architecture that integrates fast and slow thinking, with a total parameter count of 295 billion and an activated parameter count of 21 billion, supporting a long context window of 256K

Source: Zhitong The copyright of this article belongs to the original author/organization.
The views expressed herein are solely those of the author and do not reflect the stance of the platform. The content is intended for investment reference purposes only and shall not be considered as investment advice. Please contact us if you have any questions or suggestions regarding the content services provided by the platform.