Tencent unveils Hy4 model, bolstering open-source AI with advanced capabilities

Tencent Hunyuan officially launched Hy4, the latest iteration of its large language model, on November 4. This release significantly expands the model's pre-training and post-training scope compared to the preview version from late August. Notably, Hy4 introduces visual capabilities, enabling it to handle image processing, chart analysis, and mathematical visual reasoning by integrating perception, reasoning, and tool calling. Furthermore, the model supports computer operation, allowing for mixed use of command line and graphical interfaces. Tencent also trained Hy4 specifically on professional software like Blender and CAD, alongside long-horizon cross-application tasks to enhance planning. Its five primary application areas include vision and computer tasks, front-end and game development, finance and office work, scientific research, and long-term personal assistants. While Hy4 is available on Tencent Cloud and OpenRouter with no change to API pricing, Tencent is committed to the open-source community by releasing the model weights and system card, collaborating with vLLM, SGLang, and Inferact for ecosystem support. Despite its strength, the company admitted a gap remains versus top closed-source models, particularly in frontier mathematics and long-horizon development, while simultaneously announcing the upcoming launch of a cost-effective Hy4 flash version.