88Z AI, 23 April — The Tencent Hunyuan team has released the Hy3 preview model, marking their first achievement after rebuilding training infrastructure. Featuring a Mixture-of-Experts architecture that integrates fast and slow thinking, the model has 295B total parameters, 21B activated parameters, and supports a 256K context window. As the most intelligent model in the Hunyuan series, it excels in challenging reasoning benchmarks, with significantly enhanced agent and coding capabilities, enabling the one-shot g
At a glance
- Hy3 preview uses a MoE architecture integrating fast and slow thinking, marking the first result of Hunyuan's infrastructure rebuild
- The model excels in difficult reasoning benchmarks and can generate complete WeChat mini-programs in one shot
The update
The Tencent Hunyuan team has officially launched the Hy3 preview large model, representing their first major achievement following a comprehensive reconstruction of pre-training, reinforcement learning, and infrastructure. The model innovatively adopts a Mixture-of-Experts (MoE) architecture that integrates fast and slow thinking mechanisms. It features 295 billion total parameters, 21 billion activated parameters, and supports a context window of up to 256K, positioning it as the most intelligent model in the Hunyuan series to date. In terms of performance, Hy3 preview has achieved outstanding results on challenging reasoning benchmarks such as FrontierScience-Olympiad and IMOAnswerBench. Furthermore, the model's agent collaboration and code generation capabilities have been significantly enhanced, demonstrating the powerful ability to generate complete WeChat mini-programs in a single pass, showcasing its distinct advantages in handling complex tasks.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 腾讯Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.