Tencent Hunyuan T1: The 389-billion-parameter model is revolutionizing China’s AI landscape

The launch of Hunyuan-Large with 389 billion parameters marks Tencent’s ambitious move to take a leading position in the global AI race.

Chinese tech giant Tencent has reached a significant milestone with its proprietary AI model Hunyuan. Since its initial launch in September 2023, the model has evolved rapidly and is now integrated into over 50 of the company’s core products. Particularly notable is the release of Hunyuan-Large in November 2024 – a Mixture-of-Experts (MoE) model based on the Transformer architecture with an impressive 389 billion total parameters, 52 billion of which are active.

Hunyuan is characterized by particular strengths in Chinese language processing and outperforms competing models in logical reasoning and reliable task execution. An important advance: the tendency towards hallucinations – i.e. incorrect output in terms of content – has been reduced by 30-50% compared to models such as Llama 2.

Industrial application instead of academic gimmickry

Tencent’s pragmatic approach is reflected in the broad implementation of the model in various business areas. Hunyuan has already been used in more than 700 business scenarios within the company. The integration into products such as Tencent Cloud, Tencent Games, Tencent Meeting and Tencent Docs illustrates the focus on practical added value. In Tencent Meeting, the AI automatically generates meeting minutes, while in Tencent Docs it supports text creation in a wide variety of contexts.

The latest innovation is Hunyuan 3D 2.0, an advanced 3D generation system that was introduced in January 2025. It includes Hunyuan3D-DiT for shape generation and Hunyuan3D-Paint for texture synthesis. The associated Hunyuan3D Studio platform greatly simplifies 3D content creation, lowering the barriers to entry for developers and creators in industries such as gaming.

Summary:

• Tencent has developed Hunyuan-Large, an AI model with 389 billion parameters that achieves top performance on benchmarks such as MMLU and CMMLU
• The model supports context lengths up to 256K tokens and reduces hallucinations by 30-50% compared to competing models
• Integration with over 50 core products and 700 business scenarios within Tencent
• Hunyuan 3D 2.0 provides advanced 3D generation capabilities with specialized modules for shape and texture
• Enterprises in China can use Hunyuan via Tencent Cloud and customize for specific requirements

Source: Tencent