Tencent released the official version of its flagship large language model, Hunyuan Hy4, on November 4, adding vision understanding and computer-operation capabilities for the first time as the company seeks to compete with top-tier open-weight rivals1,2.
The model can process images, analyse complex charts, and conduct visual mathematical reasoning, according to Tencent. Computer-operation capabilities — the ability to interact with software interfaces autonomously — are also new to the Hunyuan line.
Compared with the August preview version, the official release expands both pre-training and post-training scale. On game-development benchmarks, Hy4's GameDev score rose 19.1% and its GameLogic score rose 9.8% over the preview. Tencent said the model's overall performance has entered the top tier of open-source models. As a demonstration of agentic capability, Tencent said Hy4 completed a 3D reconstruction of a mouse embryo comprising nearly 7 million cells in 1.5 hours without human intervention.
The model is designed to power Tencent's enterprise software ecosystem with enhanced agentic capabilities. Hy4 is now live in Yuanbao, the WeCom smart assistant, ima, and Tencent Cloud Tokenhub. API access is also available on external platforms including OpenRouter.
Tencent said the official version keeps the original API pricing, with standard API calls offered at a limited-time 50 percent discount through the end of the year. Model weights are slated for open-sourcing soon, and a cost-effective Flash version is also forthcoming.
The release lands in a competitive open-weight field. South China Morning Post framed the launch as part of a heating race among Tencent, DeepSeek, Z.ai, and Alibaba. It also arrives the same day as reports that Oracle scrapped a roughly $7 billion AI chip lease deal with Tencent under pressure from the Trump administration[1].
ANALYSIS Losing access to a large external compute lease could raise the strategic importance of efficient open-weight models for Tencent's AI roadmap.