Tencent released Hy3, a 295-billion-parameter language model, under the Apache 2.0 open-source license. The release signals China’s AI labs are willing to compete on transparency and openness, not just market dominance.

Hy3 joins a wave of Chinese models: Alibaba’s Qwen, Baidu’s Ernie, ByteDance’s models. The strategy is different from Western labs like OpenAI, which keep frontier models proprietary. Chinese companies are building network effects through open release.
Open Source Shifts the Competition
When models are open source, the value shifts from the model itself to applications, fine-tuning, and deployment infrastructure. Tencent profits when its developers build on Hy3, not from licensing the base model.
This approach also helps Tencent navigate Chinese regulations. Open models face fewer government scrutiny issues than proprietary ones. Transparency builds trust with authorities.
Chinese Models’ Market Share
According to OpenRouter data cited by CNBC, Chinese AI models accounted for more than 30 percent of weekly token usage by U.S. users in early July 2026. That’s a stunning share, driven largely by cost and performance parity with Western models.
Tencent’s open release of Hy3 accelerates the commoditization of frontier language models. Proprietary advantage is eroding fast.



