edge System One research model for computer use from trycua (the 24.9k-star cua repo, MIT-licensed source). A byte-level transformer encoder with an option-attention head: for each interface element it picks fill / check / click / skip from a fixed option set, scoring elements independently. SOURCE-ONLY RELEASE: the model card states no weights, training datasets, or checkpoint artifacts are distributed yet, and claims no performance result. The 99.7 percent vs Jev comparison circulating on X is from the vendor launch thread, not from this artifact.
- mit (source only no weights published)
- Sep 2026
Guides covering CUA-S1-FORMS #
Save your hardware and every model page answers the real question: will it run on your machine, and how fast?
Or run it in the cloud #
No per-token API provider pricing tracked for CUA-S1-FORMS yet.
For flagship list prices, see the
[calculator](https://tokenstead.ai/calculator).
Inference cost over time #
Data accumulates from the first daily sync - longer ranges populate over time. Prices come from OpenRouter snapshots, not a historical API.