One week after the official release of Tencent HuaYuan Hy3, its total call volume has seen a more than 68-fold surge compared to the previous generation Hy2, and it has topped the global large model call volume list on OpenRouter, demonstrating strong market penetration.

This model was officially unveiled on July 6, 2026. Compared to the previous preview version, Hy3 has made significant improvements in overall stability and cost-effectiveness, and its intelligence level is now comparable to flagship models with 2 to 5 times its parameter scale.

QQ20260716-092304.jpg

In terms of application ecosystem, Hy3 has been fully integrated into multiple core services such as WorkBuddy, CodeBuddy, Yuanbao, Marvis, and ima. Its API has also been launched on Tencent Cloud TokenHub and several overseas platforms. In agent products, Hy3 shows strong competitiveness. For example, in WorkBuddy, as many as 60% of users who can choose their own models have selected Hy3.

While promoting commercialization, Tencent HuaYuan is also lowering the entry barriers for the open-source community. On July 14, the team released 1-bit and 4-bit quantized versions of Hy3 based on developer needs, allowing the model, which previously required multi-card clusters, to run on a single inference graphics card or a local machine with sufficient memory.

The sharp increase in Hy3's call volume and the rapid follow-up of the open-source ecosystem not only confirm the technical achievements of Tencent in model lightweighting and engineering implementation, but also reflect that the large model industry is accelerating from simple parameter scale competition toward an application era focusing on extreme cost-effectiveness and end-side implementability.