Chinese company Tencent has launched a new open-source AI model that can operate computers, interpret complex images and handle demanding technical tasks. The model, called Hy4, marks Tencent’s latest attempt to compete with leading open-source AI systems.
Hy4 adds visual understanding to Tencent’s Hy model family for the first time. It can process detailed images, interpret complicated diagrams, and solve math problems using visual information. The model also introduces computer usage features and allows interaction with command line tools and graphical interfaces.
Tencent plans to release Hy4’s model weights and system maps in the coming days. The company says it has also built in cybersecurity and biological risk protections.
Computer control is becoming smarter
Hy4 can handle professional software including Blender and Computer Aided Design (CAD) applications. Tencent trained the model to handle complex workflows that require planning and multiple execution steps. These features could help automate tasks that traditionally require significant manual effort.
The model also supports 3D scene creation, web development and document production. Tencent reported an average review score of 83.18 on its WorkBuddy platform. Hy4 achieved a success rate of 96.15 percent on all tasks assessed.
It also ranked first in nine out of 12 scenarios tested in mainland China. WorkBuddy integrates Hy4 with WeCom, Tencent’s workplace collaboration software. The integration allows the model to access information from chats, emails, documents, calendars and meetings.
Hy4 can track project developments, identify unfinished tasks and flag upcoming deadlines. Tencent also plans to introduce additional team collaboration features through WorkBuddy’s upcoming Projects feature.
Financial and coding advances
Financial research represents another important application for Hy4. Tencent trained the model on thousands of specialized workspaces and tens of thousands of professional tasks. These exercises included fund research, securities analysis and quantitative investment strategies.
In a blind evaluation of 213 financial professionals, Hy4 achieved the highest overall score. The assessment included 579 real-world financial tasks. According to Tencent, about 70 percent of participating professionals had more than five years of industry experience.
Hy4 achieved a value of 3.04, just ahead of DeepSeek-V4.1-Flash-1017 with 2.97. Kimi K3.1 Pro and GLM-5.3 both recorded 2.68.
Software development was also given attention during training. Hy4 improved its GameDev benchmark score by 19.1 percent over Hy4 Preview. His GameLogic score increased another 9.8 percent.
The model supports development workflows with Unreal Engine, Unity and Godot. According to Tencent, these improvements extend to complex game mechanics and multiplayer implementations.
Lower costs and security measures
Tencent has also worked to reduce factual errors in Hy4’s retrieval of information. Internal evaluations showed that the hallucination rate fell from 12.95 percent to 7.2 percent. Errors related to unsupported assumptions fell from 23.1 percent to 5.6 percent.
Cybersecurity measures are another part of the publication. Tencent trained Hy4 on defensive security material, including vulnerability analysis and remediation. The training excludes offensive exploitative materials, according to the company.
Hy4 is available on Tencent Cloud TokenHub, WorkBuddy, CodeBuddy, Yuanbao and several other platforms. Its API also supports third-party services including OpenRouter and OpenCode.
Pricing remains unchanged from the Hy4 preview despite the reported performance improvements. Tencent charges around $0.90 per million input tokens and $2.70 per million output tokens. A cached input million costs about $0.045.
Tencent also plans to release Hy4 Flash soon. The smaller offering is aimed at applications where lower operating costs are most important.
