Tencent launches Hy4 open-weights model
Tencent's Hy4 brings image understanding and computer use to its open-model line for the first time, but the weights have not been released yet and most published results come from Tencent's own evaluations.
Tencent announced Hy4, the latest generation of its large language model, and says it ranks among the world's leading open-source models. The company says it will publish the model weights and system cards "in the coming days," with an emphasis on cybersecurity and biological safety. They are not available yet.
Hy4 is the first model in the line to understand images and to use computers. Tencent says it can operate both command-line and graphical interfaces, and that it was trained on professional software such as Blender and CAD tools for long-horizon tasks.
Most of the performance numbers come from Tencent's own tests:
- Finance: In a blind evaluation run with Tencent Financial Technology, 213 financial professionals rated 579 real-world tasks. Hy4 scored 3.04, ahead of DeepSeek-V4.1-Flash-1017 (2.97), Kimi K3.1 Pro (2.68) and GLM-5.3 (2.68).
- Hallucinations: In internal tests, the hallucination rate fell from 12.95% in Hy4 Preview to 7.2%. Errors caused by unsupported assumptions fell from 23.1% to 5.6%.
- Game development: Compared with Hy4 Preview, scores rose 19.1% on GameDev and 9.8% on GameLogic. Tencent cites integration with Unreal Engine, Unity and Godot.
- WorkBuddy: On Tencent's office-assistant product, Hy4 averaged a score of 83.18 with a 96.15% pass rate. Tencent says it ranked first in nine of 12 scenarios in mainland China.
Tencent also says Hy4 is moving toward "recursive self-improvement." Using its Hyra 2.0 harness, the model optimized the pre-training infrastructure for smaller models that share its architecture. Tencent has open-sourced HyCell, a model for single-cell representations, built with Hy4 and Hyra 2.0.
Tencent has not disclosed the model's parameter count, architecture or license.