Tencent Huanyuan: Hy3 preview goes live for two weeks, Token calls increase by 10 times.

On May 7th, Tencent Mixprim announced that since its launch, the token usage of Hy3 preview has been continuously increasing. Currently, the total usage is ten times more than the previous generation model Hy2. Especially, the token usage in code and smart agent scenarios has increased significantly, with a growth rate of over 16.5 times in Tencent's WorkBuddy/Codebuddy and Qclaw applications. In addition, data from OpenRouter shows that in the past week, Hy3 preview has achieved the top spot in both the weekly leaderboard and market share with a total of 3.66 trillion token usage, and also ranks first in programming and tool calling scenarios. The head of Tencent Mixprim said, "Hy3 preview focuses on practicality, and in order to collect user feedback and understand the model's performance in real scenarios, Hy3 preview was launched with a free trial activity on OpenRouter for developers to try out the model for free. During this process, we have seen a continuous increase in the model's usage, and have also received a lot of positive and negative feedback, which has provided direction for the model's future iterations. We would like to express our special thanks to developers for their attention and recognition of Mixprim." It is understood that Hy3 preview is the first model launched by Tencent Mixprim after technical restructuring, adopting a Mixed Expert (MoE) architecture combining fast and slow thinking, with a total parameter count of 295 billion and an activation parameter count of 21 billion, supporting a long context window of 256K.