Tencent has officially launched its next generation large language model called Hy4, positioning the system as one of the most capable open source releases on the market. The architecture introduces native visual understanding, automated computer operation through graphic interfaces, and targeted training for complex enterprise workflows. Tencent is making the model weights open source while deploying the technology immediately across its commercial cloud platforms.
Tencent expanded the training data and computational scale compared to the preview build. Hy4 processes fine detail images, breaks down complex charts, and performs mathematical visual reasoning. The model also operates desktop software directly. By reading command line inputs alongside graphical user interfaces, Hy4 carries out multi step tasks across desktop environments. Tencent applied dedicated training datasets for professional software like Blender and CAD, improving the system during long running cross application workflows.
Internal evaluations show Hy4 achieved an average benchmark score of 83.18 with a 96.15% task pass rate, ranking first in mainland China across 9 out of 12 enterprise test environments. The model powers feat
