DgcityTencent released its next-generation artificial intelligence model on Wednesday, calling it the first version of its Hunyuan line capable of standing alongside the leading open-source models built anywhere. The new model, named Hy4, adds image understanding for the first time and brings what the company described as major gains in following instructions, handling long documents and carrying out multi-step research tasks.
Hy4 is the finished successor to a preview version Tencent published in late August. Like many of today’s largest systems it uses a mixture-of-experts design: the preview carried 770 billion parameters in total, with about 49 billion active for any given request, and a context window stretching past one million tokens. Tencent said the full version sharply expands both early training and later fine-tuning while keeping the same price to call the model over its cloud service.
The system is already embedded across Tencent’s product lineup, including the Yuanbao chatbot, the CodeBuddy programming assistant, the WorkBuddy office tool and the ima knowledge app, with connections to the TokenHub cloud platform and the third-party OpenRouter service. The company said integrations are also being prepared for QQ, WeRead and even its flagship games, where the model can power non-player characters and in-game help.
The more notable shift is Tencent’s stance on openness. The company plans to publish Hy4’s weights in the coming days alongside a detailed system card covering cybersecurity and biological-safety testing, and it is working with the vLLM and SGLang engineering communities so the model runs smoothly on standard inference software. It is a strategic turn for a firm that long kept its models proprietary, and it lands in a year when Chinese developers including DeepSeek, Alibaba’s Qwen and Zhipu’s GLM have turned open weights into one of the country’s most effective answers to foreign technology controls.
For developers outside China, the practical effect is a wider menu of capable models that can be downloaded and modified rather than merely rented. Tencent framed the release as a bet that sharing its work will accelerate the field more than hoarding it ever could — and as an opening gambit in its effort to sell the cloud computing behind the models.
