Tencent launched Hy4, the latest flagship model in its Hunyuan line, on 4 November 2026. It is a mixture-of-experts model with 770 billion total parameters and a one-million-token context window, and the first Hy-series model that can both see images and operate a computer — putting Tencent up against rivals such as Moonshot's Kimi K3.1 Pro in the race for agentic AI.

What is new

Earlier Hunyuan models were text-only. Hy4 adds vision, so it can interpret charts, documents and screenshots, and computer use, so it can operate software via a command line or a graphical interface. Tencent says training covers professional software such as Blender and CAD tools, plus office work, finance and game development.

Price and where to try it

Hy4 is available now as a hosted service through Tencent Cloud TokenHub, WorkBuddy and CodeBuddy, Yuanbao and the WeCom assistant, and through gateways including OpenRouter. The list price is about 6 yuan per million input tokens and 18 yuan per million output tokens, with a 50 per cent discount on TokenHub until 31 December 2026, and a limited free quota in WorkBuddy and CodeBuddy from 4 to 15 November.

The open-weights question

Despite the open-source framing in some coverage, Hy4's full weights have not been released. Tencent says it plans to publish them, with support from vLLM and SGLang and a fuller system card, but it has given no date, and only the August preview weights are on Hugging Face under Apache 2.0. Until the weights ship, Hy4 is an API model with a promise attached. Its benchmark claims come from Tencent's own tests, including a finance evaluation of 579 tasks scored by 213 professionals, in which Hy4 scored 3.04 against 2.97 for DeepSeek-V4.1-Flash. Independent testing has not yet had time to appear.