Tencent releases first Hunyuan large model
Unveiled at Tencent's Global Digital Ecosystem Summit, the mixture-of-experts model exceeded 100 billion parameters and launched for enterprise access via Tencent Cloud, not consumers.
- Models & capabilities
- Minor
Tencent unveiled Hunyuan, its first general-purpose large language model, at the company’s Global Digital Ecosystem Summit. The model exceeded 100 billion parameters, used a mixture-of-experts architecture built on the Transformer, and was reported to have been trained on roughly 2 trillion tokens of pretraining data, with Tencent citing particular strength in Chinese-language comprehension, content creation, logical reasoning and task execution.
Rather than a consumer chatbot launch, Tencent made Hunyuan available first through Tencent Cloud’s model-as-a-service offering, giving enterprise customers API access to build applications on top of it, and introduced a companion Hunyuan Bot as a WeChat mini-program. This enterprise-first approach differed from the consumer-facing launches Baidu and ByteDance had made with Ernie Bot and Doubao in the same window, reflecting Tencent’s position as a cloud and enterprise-software provider as much as a consumer-facing developer of AI products.
The release marked Tencent’s formal entry into China’s large-model race, alongside Baidu, Alibaba, ByteDance and others, each racing to ship a foundation model over the course of 2023. Hunyuan became the base for a subsequent series of updates and specialised model releases from Tencent, including later multimodal and video-generation variants built on the same foundation.