Tencent has released Hy3, an Apache-licensed open model that achieves remarkable performance benchmarks while operating at roughly half the size of competitor models. The release marks a significant shift in the Chinese AI landscape toward more accessible, developer-friendly licensing.
The Open Model Advantage
Unlike many Chinese AI releases that come with restrictive licenses, Hy3 is fully Apache-licensed — meaning developers can freely use, modify, and commercialize the model without royalty concerns. This positions it as a serious alternative for enterprise IT teams evaluating open models, particularly those operating under export compliance constraints.
Performance Benchmarks
According to benchmarks published by Tencent, Hy3 outperforms Zhipu’s GLM-5.2 across nearly every category — except coding. The model is specifically sized to run on export-compliant GPUs, addressing a critical need for organizations restricted from accessing the most powerful AI chips.
The half-size architecture delivers a significant hallucination rate reduction compared to larger competitors, making it particularly suitable for production deployments where reliability matters.
Strategic Implications
Hy3’s release represents a broader trend in the Chinese AI ecosystem: competing on efficiency rather than pure scale. With US export controls limiting access to high-end GPUs, Chinese labs are increasingly focusing on model optimization and distillation techniques to deliver comparable results with limited compute.
For developers and enterprises, the model offers a compelling option — particularly those seeking an open, commercially safe alternative to heavily restricted frontier models. The Apache license eliminates legal ambiguity around deployment, a persistent concern with many AI tools originating from different regulatory jurisdictions.
The model is available now for macOS, Windows, and Linux, with bring-your-own-key configurations supporting third-party model integration.