China’s Tencent has released the full version of its Hy4 large language model, adding visual understanding and computer-use functions as Chinese developers compete to expand open-model capabilities for enterprise and consumer applications.
The November 4 release follows the Hy4 preview that Tencent introduced in August. Tencent said the new version uses expanded pretraining and post-training data and can process images, reason over charts, and handle mathematical visual tasks. It also supports computer use through command-line and graphical interfaces.
TNGlobal reported that the preview had 770 billion total parameters, 49 billion active parameters, and a context window of more than 1 million tokens. Tencent did not disclose a revised parameter count or context limit for the full release. The company said it would publish the full model’s weights and system card later, making their availability an announced next step rather than a completed part of the launch.
According to Tencent’s announcement, the company conducted specialized training for professional software including Blender and computer-aided design applications, as well as longer tasks that require work across multiple applications. Tencent did not provide independently verified measures of those computer-use capabilities.
The company said Hy4 is available through Tencent Cloud TokenHub, WorkBuddy, CodeBuddy, Yuanbao, WeCom’s AI assistant, and ima. Third-party application programming interface access is available through OpenRouter, Hermes, and OpenCode, while Tencent is integrating the model into additional products including QQ, Honor of Kings, Game of Peace, WeRead, and Tencent News.
Tencent reported that 213 finance professionals evaluated Hy4 on 579 tasks, with 153 participants coming from external financial institutions. The company said Hy4 received an average score of 3.04, compared with 2.97 for DeepSeek-V4.1-Flash-1017 and 2.68 each for Kimi K3.1 Pro and GLM-5.3. These results were organized or reported by Tencent and have not been independently verified.
The release also emphasizes cybersecurity and biological-safety controls. Tencent said it trained Hy4 with defensive cybersecurity material related to proof-of-concept generation and vulnerability remediation while excluding offensive material on exploitation, reverse engineering, and penetration testing.
Internal specialists from Tencent’s Xuanwu Lab and Keen Security Lab assessed the model’s capabilities in areas including cryptographic analysis, binary deobfuscation, and vulnerability exploitation. Tencent said the tests did not identify elevated offensive risks. Reuters reported that the safety findings were based on the company’s internal testing and that Tencent plans to publish the model weights and a report describing its safety assessments.
Tencent said Hy4 retains the preview version’s pricing at CNY 6 per million input tokens, CNY 18 per million output tokens, and CNY 0.30 per million cached input tokens. It also plans to release a lower-cost Hy4 Flash version.