Hy4 is Tencent’s latest flagship model, aimed at office tasks, financial analysis and game development. New since the preview, per Tencent: visual understanding for charts and images, and computer use across command line and graphical interfaces, with training on Blender and CAD. Tencent says the internal hallucination rate fell from 12.95% to 7.2%. Available on Tencent Cloud TokenHub, OpenRouter and Tencent apps. Weights are promised in the coming days.

Hy4 Preview is a 770B MoE model (49B active, 1M context) from Tencent, built for long-horizon agentic tasks. It autonomously handles coding, game dev, and complex document analysis, running its own tests and fixing bugs before delivery.

Hy3 Preview is a 295B-parameter MoE model (21B active) from Tencent. Built on a completely rebuilt RL infrastructure, it delivers best-in-class cost efficiency and significant improvements in complex reasoning, instruction following, and agentic coding tasks.

Hunyuan-A13B is Tencent's new open-source MoE model (13B active parameters). It delivers top-tier performance with low computational cost, supports a 256K context window, and includes a thinking mode.