I noted Tencent’s Hy4 as a multimodal model with computer-use capabilities, while treating its benchmark figures as Tencent’s own reports and its promised weights and system card as forthcoming.
Hy4 is Tencent's new flagship model, the first in its Hunyuan line with vision and computer use built in.
Hy4 reads charts and screenshots and drives command-line and graphical interfaces, including professional tools such as Blender and CAD software.
On Tencent's own tests, Hy4 scored 78.9 on OSWorld 2.1, close behind Claude Opus 5.5 at 81.8.
Hy4 scored 3.04 out of 4 across 579 real finance tasks.
Its hallucination rate in Tencent's finance test fell from 15.76% to 5.45%.
I noted the caveat that these are Tencent's own numbers; the model weights are not out yet, and a system card is promised to follow.
I asked: Are open models closing the gap on the closed frontier or just catching up on benchmarks that suit them?