Tencent releases Hy4, adding vision and computer use, with open weights promised
Tencent's new flagship model is available now through its own products and OpenRouter; the company says the weights and a safety system card will follow.
By 1EyeAI Desk ·
Tencent on Nov. 4 released Hy4, the latest generation of its Hy large language model. The company says the model adds visual understanding for the first time, including reading charts and documents and reasoning across multiple images, as well as computer use: operating command-line tools and graphical interfaces. TechNode reports the training targeted applications such as Blender and CAD software.
Hy4 is available on Tencent Cloud and in Tencent products including CodeBuddy, WorkBuddy, Yuanbao and ima, and through third-party platforms including OpenRouter. Tencent's Hy research page lists API pricing of $0.834 per million input tokens, $2.501 per million output tokens and $0.042 per million cached input tokens. That's the same as the Hy4 preview. Tencent Cloud TokenHub is offering a 50% discount through Dec. 31.
Tencent says it plans to open-source the Hy4 weights "in the coming days" and to publish a system card with its safety evaluations at the same time. Reuters reports the model was trained with limited cybersecurity data to reduce its potential for offensive hacking use. Tencent's announcement says the open-source release will focus on cybersecurity and biological safety.
TechNode reports the model has 770 billion parameters. Tencent's Hugging Face page for the earlier Hy4 preview describes a 770B-parameter mixture-of-experts model with 49B parameters active per token and a 1M-token context window. The specifications for the final release have not yet been published.