Tencent Hy4 Released: Adds Vision Capabilities, Open Models Begin Competing on Scenarios

The Tencent Hy team has released Hy4, adding vision capabilities in this version. According to Tencent Hy on X, Hy4 is now among the leading open models and improves productivity in scenarios such as finance, frontend, gaming, office work, and research. The company also emphasized that cybersecurity and biosecurity were considered during model construction.

This message, forwarded by Eric Hartford, did not elaborate on parameters, licensing, or benchmark details. What can be confirmed is the vision capability, the list of scenarios, and the safety statement. For people watching open models, vision entering the main release is not surprising, but it directly affects model selection: previously, a text model had to be paired with a bolt-on vision pipeline; now one can see whether a single model can handle inputs such as screenshots, interfaces, and document images.

The official list of scenarios is broad. Finance, frontend, gaming, office work, and research differ in error tolerance, data sensitivity, and interaction patterns. Frontend and gaming depend more on screenshot understanding and UI reconstruction, while finance and office work depend more on fields, tables, and workflows in document images. If future evaluations provide only a single overall score, its reference value will be limited; it is more practical to look at results by scenario.

Try it here: https://hy.tencent.ai/research/hy4?langVersion=en

Original tweet link: https://x.com/QuixiAI/status/2117950628546625850