TENCENT (00700.HK) released the preview version of the HY3 model last Thursday (23rd), marking its first model since rebuilding its pre-training and r...
Model details
TENCENT (00700.HK) released the preview version of the HY3 model last Thursday (23rd), marking its first model since rebuilding its pre-training and r...
Tencent unveils Hy3 preview — a 295B-parameter MoE model with 256K context. First major release since the Hunyuan pipeline rebuild, priced at RMB 1.2/M input tokens.
Tencent is launching Hy3 preview with the usual benchmark claims expected of a new large language model. But the more distinctive part of the rollout is where
腾讯于1998年11月成立,是一家互联网公司,通过技术丰富互联网用户的生活,助力企业数字化升级。我们的使命是“用户为本 科技向善”。Founded in 1998, Tencent is an Internet-based platform company using technology to enrich the lives of Internet users and assist the digital upgrade of enterprises. Our mission is "Value for Users,
The news blog specialized in Japanese culture, odd news, gadgets and all other funny stuffs. Updated everyday.
Tencent just open-sourced Hy3 preview, a new model that punches above its weight on coding agents, reasoning, and search—built in under three months., Predict, earn, and engage with Myriad Markets interactive prediction markets on your favorite platforms and news sources., Predict, earn, and engage with Myriad Markets
Keep Reviews Moving
When AI speeds up shipping, review queues get exposed fast. CodeRabbit reviews pull requests quickly, catches issues that surface late, and adds coverage before code reaches production.
Developers already feel this
The Hy3 preview represents a significant architectural shift for the Hunyuan family, utilizing a mixture-of-experts design that fuses fast-and-slow-thinking capabilities. By employing 295 billion total parameters while activating only 21 billion during inference, the model balances high-level intelligence with efficient computational throughput. This design intent focuses on delivering robust performance for intricate reasoning, precise instruction following, and complex agentic workflows, positioning the model as a versatile tool for developers tackling demanding technical and creative tasks.
Emerging from a comprehensive rebuild of the company's pre-training and reinforcement-learning infrastructure, this model reflects a new development trajectory under updated leadership. The transition to this refined pipeline allowed for a leaner, more powerful architecture compared to its predecessors, despite a reduction in total parameter count. With its focus on in-context learning and advanced coding proficiency, the model is engineered to serve as a foundational asset for real-world applications, signaling a forward-looking commitment to enhancing practical usability in AI-driven environments.
Why teams adopt it
Discuss this model
Add corrections, implementation notes, pricing changes, or usage caveats for other readers.