Tencent Releases Hy3 AI Model Under Permissive Apache 2.0 License
Tencent has released its Hy3 AI model under the open-source Apache 2.0 license, removing previous geographic restrictions and enabling wider enterprise adoption.
Chinese technology giant Tencent has fully released its Hy3 artificial intelligence model under the permissive Apache 2.0 license. This move removes previous licensing barriers that had excluded companies in regions like the European Union, the United Kingdom, and South Korea from using the model.
Hy3, built on a 295-billion-parameter Mixture-of-Experts (MoE) architecture, is now readily available for enterprise deployment. Tencent positions the model as competitive with larger models, making it a significant contender in the open-source AI landscape.
Refined through feedback from over 50 internal product teams following its preview release, Hy3 boasts improved reliability and production economics. The company highlights these enhancements as crucial for real-world application.
According to Tencent's evaluations, Hy3 excels in tasks like search and tool orchestration, rivaling even much larger models. However, the model lags behind Zhipu AI's GLM-5.2 in coding benchmarks, despite being considerably smaller.
Key improvements in Hy3 include a significant reduction in hallucination rates and error rates in internal assessments. Tencent also emphasizes the model's consistent performance across various deployment scaffolds, addressing a critical need for production environments.