Tencent, a prominent Chinese technology company, has introduced its own large language model (LLM) called "Hunyuan," which bears similarities to OpenAI's ChatGPT. Hunyuan marks Tencent's entry into the realm of "base models," designed as a foundational framework upon w various AI APIs can operate.
Hunyuan represents a comprehensive suite of AI tools, as outlined in a Tencent blog post: "Tencent's basic model supports a wide range of functions, covering image creation, copywriting, text recognition, customer service, etc. These will play an important role in key industries such as finance, public services, social media, e-commerce, transportation, games, etc."
The AI system is seamlessly integrated into Tencent's extensive application and service ecosystem, bridging Tencent Cloud, Tencent Marketing Solutions, Tencent Games, Tencent Fintech Services, Tencent Meetings, Tencent Documents, WeChat Search, and QQ Browser. This launch comes amidst continuing frosty relations between the US and China following the Biden administration's October 2022 export ban on certain types of computer chips, including hardware essential for the development and training of AI systems.
Tencent emphasized that the introduction of Hunan underscores its "commitment to open cooperation within the ecosystem," enabling domestic companies to access high-quality model services while international enterprises can leverage Tencent to enter the Chinese market. Currently, ChatGPT is not accessible While several LLMs have previously entered the Chinese market, such as Alibaba and Baidu, Hunyuan's association with Tencent, China's largest technology company with nationwide reach and the popularity of WeChat and related apps, positions it uniquely to make a significant impact.
In terms of raw data, Hunyuan's capabilities and functions are reportedly similar to GPT-3 (OpenAI's core model circa 2022), boasting 100 billion parameters and 2 trillion tokens, making it one of the world's most powerful LLMs. A notable distinction is that Hunyuan has been trained on a substantial Chinese text corpus, potentially granting it an advantage for operating effectively within Chinese-language environments, compared to models primarily trained on non-Chinese text.
