Introducing CUA-S1
Cua's X announcement of CUA-S1, a family of small open System One models specialised for computer use, starting with form filling.
This summary is based on how curated lists describe the project; its README was not available to review.
Overview
In this X post, Cua, a company building computer-use agent infrastructure, announces CUA-S1: an open-source family of small, specialised System One models for computer use. The first specialist targets form filling, and the post asks the community which specialist should be trained next.
It matters for the decision-model ecosystem because it applies the System One idea, fast typed decisions rather than generated text, to narrow computer-use skills that a larger agent can call. Only the list summary was available, so model sizes, licence, benchmarks and the question format used are not documented here.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which form field or value should be filled next?(inferred) | — |
How it works
- Agent encounters a form on screen
- Specialist System One model makes the typed decision for the field
- Agent executes the action and continues
Strengths
- Small specialised models can be cheap and fast for repeated computer-use steps
- Open-source release allows local use
Caveats
- Announcement only; details not in the source list
- Only the form-filling specialist is mentioned
At a glance
- Type
- Discussion
- Status
- discussion
Use cases and integrations
How curated lists describe it
X: Cua open-sources a family of small, specialised System One models for computer use, starting with form filling and asking what the next specialist should learn.
yibie/awesome-jev
Frequently asked questions
What is CUA-S1?
A family of small open System One models specialised for computer-use tasks, starting with form filling.
Who made it?
Cua, per their X announcement.
Where are the weights?
The list summary does not link them; check Cua's announcement.
Related projects
Computer-use platform (Cua Spaces, Driver, Lume) with CUA-S1 small decision models and a preview jev-use example for bounded actions.
ProjectbetaPythonRuns Intern-Decision-0.8B on Apple Silicon behind a local /v1/systemone endpoint that accepts screenshots for agent decisions.
ServicebetaPythonFrom-scratch research reconstruction of a small System-One decision model, used to route and gate commands in a Windows palette.
ProjectresearchPythonOneJev: open multimodal System One model (0.8B-27B) answering typed questions about screenshots, photos, video and text in one pass.
ModelbetaPython- CLM★ 3,014
CLM-8B, a contrastively trained System One model that embeds states and actions separately, served behind a TypeSafe-compatible API.
ServiceresearchPython 706K-parameter Jev-style option scorer that rates fill, check, click or skip for each form element in one parallel pass.
ModelresearchPython