Tencent Hunyuan Big Model Official Website Launched, Now Open to the Public


Tencent Cloud announced that the free public testing period for the Hy3 preview and DeepSeek-V4-Pro models in its intelligent agent development platform will end on May 27, 2026 at 10:00. These two models have received widespread attention during the public testing period, providing powerful intelligent solutions to help developers and enterprises improve efficiency and business capabilities. Tencent Cloud stated that after long-term debugging and testing, the models now have higher stability and intelligence levels.
Tencent Cloud announced that it will cease support for the three older models, DeepSeek-V3-0324, DeepSeek-V3.1-Terminus, and DeepSeek-R1-0528, starting from May 22, 2026. Users are advised to switch to the new version promptly to ensure continuous and stable service.
Baidu released its new language model Ernie5.1 on May 11, 2026, based on the pre-trained foundation of Ernie5.0 with 2.4 trillion parameters. Through a 'one-time elastic training framework', it achieves single training optimization for multiple model sizes, with pre-training cost only 6% of similar models. As of May 9, the model ranked fourth globally and first in China on the Arena Search ranking with 1223 points, demonstrating high resource utilization and performance balance.
Sichuan Hongmofang and Tencent Cloud signed a cooperation agreement in Chongqing, jointly developing the next generation of AI dolls. Both parties will integrate the advantages of cloud computing, AI large models, and smart hardware to promote the transformation of AI toys from functional tools to emotional companions, achieving a combination of software and hardware, and endowing traditional toys with stronger interactive and companion capabilities.
The Tencent Cloud Agent platform QClaw released version v0.2.14, which is the largest update to date. The upgrade includes integrating the Hermes framework, supporting the creation and operation of Hermes-type Agents, achieving diversified underlying models, significantly lowering the AI usage threshold, and allowing users to schedule multiple models within a single application.