Hy4 adds vision and computer use. How reliable will it be on long tasks?
Tencent just dropped Hy4. Vision and computer use are probably the two biggest new additions. For the first time, Hy can understand images and complex charts. It can also work…
By Charlotte Hayes
Tencent just dropped Hy4. Vision and computer use are probably the two biggest new additions. For the first time, Hy can understand images and complex charts. It can also work across CLI and GUI, with specific training for tools like Blender and CAD. The computer use part sounds pretty interesting, but there’s more to it than just clicking buttons and running commands. What happens when a button moves, a permission pop-up gets in the way, or a command fails halfway through a task? Can the model figure out what went wrong and get back on track? And what about the extra cost? Every failed action, retry, or new screenshot could mean more tokens and API calls. Over a long workflow, those costs could add up pretty quickly. There’s also the problem of keeping track of everything across a long workflow. An agent might complete several steps correctly, then lose track of its progress or …
It is part of a fast-moving week in AI, as researchers, builders, and enterprises keep adjusting to what the latest capabilities mean for them.
