Jev with Cua Driver passes 8 of 9 computer-use tasks in first test

Francesco @francedotOriginally posted 2026.09.17

Open original on X

curious how far a model built for structured decisions can go at computer use. we tried @typesafeai's jev with Cua Driver: 8/9 executable tasks passed vs 9/9 for Astra/Codex, with median decisions 13.6x faster and median task time 27% lower. small test but it makes me wonder how much of computer use becomes a general decision problem once the model has a usable representation of the UI not bad for a first public release. great work @CompleteSkeptic and team!

This post quotes another post: x.com/injaneity/status/2100177865958682860