Newsletter Subscribe
Enter your email address below and subscribe to our newsletter
[forminator_form id="25163"]

phemex+1phemex+1finance.biggo+1Tencent on Monday officially launched Hunyuan Hy3, the full release of its next-generation large language model, iterating on the preview version introduced in April and positioning the Chinese tech giant as an increasingly competitive force in open-source AI.
Hy3 is a 295-billion-parameter Mixture-of-Experts model with 21 billion activated parameters and support for a 256K token context window, released under the permissive Apache 2.0 open-source license. Despite its relatively compact activation footprint, Tencent says the model rivals flagship open-source models with two to five times more parameters. The hallucination rate dropped by more than half compared to the preview version, reaching 5.4%, while the model achieves a 90% agent task resolution rate.phemex+3
Pricing has been reduced further from the preview's RMB 1.2 per million input tokens to RMB 1 per million input tokens, with output at RMB 4 per million tokens. The model is available through Tencent Cloud's TokenHub platform and on GitHub and Hugging Face.tencent+1
The launch follows a rapid development cycle. Tencent restructured its Hunyuan team under chief AI scientist Yao Shunyu in late 2025, and the rebuilt infrastructure produced the Hy3 preview in under three months, launching it on April 23. That preview version quickly gained traction, topping the OpenRouter API aggregation platform's weekly call volume chart in early May with 3.66 trillion tokens of usage.36kr+2
The full Hy3 release builds on those gains with improved post-training data quality and expanded reinforcement learning computation. VentureBeat reported Monday that Hy3 outperforms Zhipu's GLM-5.2 across most benchmarks despite being roughly half the size.venturebeat+1
Hy3 has been integrated across Tencent's ecosystem, including WorkBuddy, CodeBuddy, Yuanbao, and QQ. Tencent's Hong Kong-listed shares rose more than 4% on the day of the announcement.finance.biggo+2
The release intensifies competition among Chinese AI labs, where open-source models from DeepSeek, Alibaba's Qwen, and Zhipu have been vying for developer adoption. With its Apache 2.0 license and aggressive pricing, Tencent is betting that accessibility and integration across its vast product suite will drive Hy3 toward becoming a default infrastructure layer for enterprise AI in China.