frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Show HN: CUA-S1 – A System One Model for Computer Use

https://github.com/trycua/cua
14•frabonacci•3h ago
Hello HN! We're Dillon and Francesco from Cua.

We were wondering how many computer use tasks actually need a full general purpose LLM (e.g. gpt-6-astra, claude-opus-5 etc.) to think through all their decisions and steps. Some tasks require thinking about a plan, exploring different paths, recovering from failure. Other tasks are a question of making local decisions, like this value should go in this box, or should I check this box, or this element should be ignored.

We wondered how far we could go with a small model trained to only make these kinds of decisions.

Our inspiration was Typesafe's Jev and its System One Model framing. This is a nod to the dichotomy between thinking quickly, automatically, and intuitively (system 1) vs. thinking slowly, analytically (system 2), as described by Daniel Kahneman.

The interesting question for us was: what happens if you give a model an interface of current context, and a set of possible choices, and you ask it to return a probability for each choice? This kind of model does not generate output token by token like most LLMs do, but rather scores the options you give it, which you can check, trust, and use to drive your app's behavior.

CUA-S1 is our answer for narrow, specialized decision models for computer use. Our first release is CUA-S1-FORMS. We built this from ideas and code in jevlike, and then trained a second model just to handle form interactions. It has 706k parameters, and the original checkpoint is 2.8 MB.

The first training iteration took less than 30 minutes on synthetic data. Given a set of structured elements and values extracted from a document, it predicts whether to use the given value, CHECK, CLICK, or SKIP for each element. It does not predict new values for text fields, and does not consider screenshots. Element decisions are scored together, and your code can order the actions, and Cua Driver will execute them one at a time.

A first evaluation of this specialist vs. hosted Jev on our form task:

- For the whole decision set: 99.7% correct vs 83.6%.

- For the subset of steps that require an action: 100% correct vs 96%.

- For the subset of steps that are just leaving already-filled fields alone: 100% correct vs 74%.

The specialist was trained specifically for this task and convention (just press skip for already filled boxes), while hosted Jev has not been fine-tuned for it, so this is an experiment in scoped specialization.

We measured 7-9 ms to score a form locally vs. 260-280 ms per call to hosted Jev including network latency, though those samples measure different things and are not end-to-end form completion times.

Our interest here is in the space between a brittle script and a general agent loop. The content and layout of form fields vary enough that scripts get unwieldy, but the set of available decisions can remain narrow and well scoped. We want to explore the possibility of a general agent encountering something novel, and passing well understood decisions over to specialists like this.

That is a direction we are looking into. The current release is for forms only. We're open sourced the synthetic data generation, training, evaluation, and Driver integration under libs/cua-s1 with an MIT license.

Comments welcome! Especially if you are building computer-use agents and have run into a recurring decision that is too variable to script but is too narrow to call another LLM for.

Comments

badatnames•7m ago
Please give us a hyper general cookie consent popup dismissal add-on :)

I built non-autoregressive decision models with RL a year ago

https://laya.convaiinnovations.com/
809•nandakishor_ml•8h ago•197 comments

AI-generated posters don’t have to be horrible

https://john.hartnup.uk/2026/06/07/ai-event-posters.html
1028•ereiamjh•10h ago•590 comments

Btrfs/ZFS/bcachefs under workloads classic benchmarks skip

https://bartosz.fenski.pl/modern-fs-benchmark/
27•farlight•1h ago•23 comments

Human brain is two separate organs, Stanford Medicine-led research finds

https://med.stanford.edu/news/all-news/2026/09/two-separate-brains.html
569•emigre•13h ago•204 comments

A graphical desktop for the ZX Spectrum

https://github.com/mindbox77/zxdesk
113•graemep•5h ago•89 comments

Tin: full-text search for Postgres

https://planetscale.com/blog/introducing-tin
132•ksec•5h ago•58 comments

Suzanne Ciani's Buchla Cookbook

https://echo.orpheusinstituut.be/article/suzannes-buchla-cookbook
18•stuart78•2d ago•4 comments

Show HN: CUA-S1 – A System One Model for Computer Use

https://github.com/trycua/cua
14•frabonacci•3h ago•1 comments

The Secret Life of Circuits

https://blog.coredump.cx/p/the-secret-life-of-circuits-is-here
240•surprisetalk•3d ago•61 comments

Android 17 is the first since 3.x to add new APIs without releasing to the AOSP

https://grapheneos.social/@GrapheneOS/117282080803799576
1050•theanonymousone•1d ago•608 comments

Supabase (YC S20) Is Hiring for OrioleDB

https://supabase.link/orioledbjob
1•awalias•2h ago

New evidence for hidden chambers beyond Tutankhamun's tomb

https://www.nature.com/articles/d41586-026-02621-2
63•rndsignals•2d ago•19 comments

Black Holes or Black Hole Stars? Astronomers Spar over 'Little Red Dots'

https://www.quantamagazine.org/black-holes-or-black-hole-stars-astronomers-spar-over-webb-telesco...
71•jandrewrogers•1d ago•29 comments

UFO Series Home Page: "UFO" TV Series from 1970

https://ufoseries.com/
5•DropDead•1d ago•1 comments

Almost Never Use AI to Write Anything Substantive

https://erichgrunewald.substack.com/p/why-you-should-almost-never-use-ai
92•erwald•3h ago•50 comments

GPT-6 Astra Solves a WWI German Radio Cipher

https://www.prinzai.com/p/gpt-6-astra-solves-a-wwi-german-radio
297•nsoonhui•12h ago•150 comments

San Francisco Onion Futures Company

https://onionfutures.com/
324•z-mach9•15h ago•139 comments

Cloudflare Quick Tunnels

https://try.cloudflare.com/
792•jcbhmr•1d ago•303 comments

How to Write with an LLM

https://sockpuppet.org/blog/2026/09/17/how-to-write-with-an-llm/
587•joeriddles•1d ago•367 comments

Adventures in Microcontroller Circuit Debugging

https://www.bigmessowires.com/2026/08/30/adventures-in-microcontroller-circuit-debugging/
12•lateatdesk•2d ago•1 comments

If math is more than proof, we need to better celebrate the rest of it

https://terrytao.wordpress.com/2026/09/18/if-math-is-more-than-proof-we-need-to-better-celebrate-...
261•num42•13h ago•220 comments

What Zig felt like, coming from Rust

https://besok.github.io/posts/what-zig-felt-like-coming-from-rust/
142•ksec•5h ago•165 comments

You can run Git on object storage if you re-make packfiles

https://www.tigrisdata.com/blog/objgit-packfiles/
120•evacchi•3d ago•30 comments

Asking Authors About Their Own Papers

https://medium.com/@TmlrOrg/asking-authors-about-their-own-papers-3d2e04e5dee0
99•stefanpie•2d ago•56 comments

Saving another 100TB of RAM

https://blog.cloudflare.com/saving-100-tb-of-ram-with-math/
439•f311a•1d ago•98 comments

Communication by means of modulated Johnson noise

https://www.pnas.org/doi/10.1073/pnas.2201337119
50•austinallegro•2d ago•25 comments

SDCC – Small Device C Compiler

https://sdcc.sourceforge.net/
118•lioeters•17h ago•28 comments

How OpenAI Used Its Own LLMs to Design Its Jalapeño Chip

https://spectrum.ieee.org/llms-for-chip-design
188•maxall4•20h ago•126 comments

Science Is Open Software

https://jepedersen.dk/blog/202505_research/
145•jegp•17h ago•64 comments

The first new cat species discovered in 100 years

https://www.nationalgeographic.com/animals/article/meet-the-first-new-cat-species-discovered-in-1...
347•ohjeez•2d ago•133 comments