Tencent has released the full version of Hy3, a 295-billion-parameter Mixture-of-Experts (MoE) model with 21 billion active parameters, under the permissive Apache 2.0 license — a reversal from April's preview release. The move removes licensing restrictions that previously blocked enterprises in the European Union, United Kingdom, and South Korea from deploying the model, addressing a key obstacle in the open-weight model market.
The model outperforms GLM-5.2 on most benchmarks despite being half its size, with the exception of coding tasks where GLM-5.2 retains an edge. For IT teams evaluating open models, the trade-offs are unusually explicit: Hy3 offers strong performance with permissive licensing, but falls short in code generation. Tencent says the model will be free on OpenRouter for two weeks.
The open-model community reacted swiftly, with researchers on X singling out the license change as the real headline. One widely shared post argued that if the scores hold up, Tencent has just become one of the leaders of open source. The Apache 2.0 license eliminates legal friction that had kept many enterprises from adopting strong Chinese open-weight models.
This release signals a broader trend: major Chinese tech firms are increasingly competing on openness rather than just raw performance. By offering a permissively licensed model that rivals proprietary alternatives, Tencent may pressure other model developers to follow suit. The move could reshape enterprise adoption patterns, particularly in regions previously locked out by restrictive licensing.
Tencent's Hunyuan team developed Hy3, though founder commentary was not included in today's announcement. The reaction from the open-source community suggests the license change may prove more consequential than the model's benchmark scores.