Complete AI Training
Sign inGet my AI kit

Your job's AI kit

Get your AI kit

Tell us who you are and what you do. We show you your kit right away and email you the link: skills, prompts, AI agents, MCP servers and courses for your job.

500+ jobs ready, and we make a kit for any other job. No payment needed to look.

Share

AI news ·

Tencent releases smaller flagship AI model under former OpenAI researcher's leadership

Tencent released Hy3, a 295-billion-parameter AI model, smaller than its 400B predecessor. It matches top Chinese rivals but lags behind OpenAI and Google DeepMind.

1 min read
IT and development

Tencent releases flagship AI model with smaller parameter count

Tencent Holdings released a new flagship AI model on Thursday, marking the first major release since former OpenAI researcher Yao Shunyu joined the company to lead foundational AI development. The open-source model, called Hy3 preview, performs on par with top Chinese competitors but still trails US leaders like OpenAI and Google DeepMind.

The model contains 295 billion parameters-a notable departure from industry trends toward trillion-parameter systems. Parameters are the mathematical variables that encode a model's capabilities and roughly correlate with computational requirements for training and deployment.

Tencent's previous flagship, HY 2.0, released in December 2025, contained over 400 billion parameters. The smaller size of Hy3 reflects a design philosophy centered on real-world business use rather than raw scale.

Built for production deployment

Tencent developed Hy3 through collaboration between its Hunyuan foundational model team and Yuanbao AI application team. The company said the model bridges "the gap between model capability and user value" by aligning technical capabilities with product requirements.

Hy3 is already deployed in Tencent's consumer app Yuanbao and CodeBuddy, a coding assistant for IT and development professionals.

Agentic capabilities see major improvements

Agentic features-which enable models to plan and execute tasks autonomously-represent one of the most significant improvements in this release. Tencent tested the model against in-house benchmarks designed to measure compatibility with OpenClaw, a popular agentic AI tool.

For developers evaluating AI infrastructure, the smaller parameter count combined with production-focused design offers a different tradeoff than maximizing model size. Understanding these design choices matters for generative AI and LLM development decisions.

Share