Tencent launches open Hy4, betting on image understanding and long context

Tencent announced the launch of the Hy4 model, a new large language model that adds image understanding capability and improves instruction following, long-context comprehension and extended agentic tasks.

The company said the model is among the world's leading open-source models, and that it intends to publish its weights and system cards in the coming days.

The Decoder website reported that the model ranked second among Chinese models in independent tests, but trails leading models such as Claude Opus 5.5 in hard mathematics and programming.

The launch reflects the acceleration of Chinese companies in releasing open models, as weights allow technical teams to run the model locally or modify it instead of relying entirely on cloud services.

What do these terms mean?

Open-source model: A model whose weights or code are published for everyone to inspect, modify and use.

Model weights: The files containing the trained values that allow the model to be run or reused.

Long context: The model's ability to process large texts and retain information across long distances in the text.

Agentic tasks: Tasks the AI performs in successive steps instead of a single short answer.

Image understanding: The model's ability to analyse images, extract information from them and link it to text.

System card: A document describing how the model was built, its limits and its intended uses.