Smaller, latency-optimised Cloudflare decision model that returns probability-scored classifications of text or images
CUA-S1-FORMS
A 2.8 MB model that chooses form field values on the local machine, with no API call
About CUA-S1-FORMS
One narrow decision is all CUA-S1-FORMS makes: which value belongs in which form field, and the model that makes it is very small. Published on 18 September 2026 by Cua, a San Francisco company, it opens a family the company calls CUA-S1, or System One: small, specialized models for the bounded choices an agent keeps having to make as it operates a computer.
How the model decides
Nothing it produces is free text. The input is a form's context plus values already pulled from documents, and in one forward pass the model scores each allowed option and returns a probability for it. That keeps the decision inside ordinary code instead of becoming one more call to a large model. It runs on the same machine as the rest of the work, so there is no API call and no per-token price.
Size, license and design
The checkpoint is 2.8 MB with 706,048 trainable parameters. It is licensed under MIT, and the weights come with the training code and the synthetic dataset. Architecturally it works on bytes: a transformer encoder with two layers, width 128 and four attention heads, which reads 224 bytes of form context and 96 bytes for each option. Training used 10,000 synthetic episodes of forms with between two and sixteen fields, divided so that no form layout appears on both the training side and the test side.
Results and how to read them
Cua reports 99.95% top-1 accuracy on its synthetic test and 100% on 196 decisions drawn from three real forms and three PDFs. A control with shuffled context scored 37%, which Cua uses to show the model reads the elements rather than guessing. Cua also publishes a head-to-head result in which it scores 99.7% and a hosted general model scores 83.6%. These are Cua's own evaluations and the real-form set is small. An independent write-up notes that the comparison favors the specialist, because Cua trained its model to treat already-completed fields as no-ops and the other model was not.
Limits stated on the model card
Its choices are limited to values already supplied as labeled entities, so it cannot work out or assemble anything absent. The training data was synthetic and has had little validation on real forms, and the encoder is centered on English.
TrustList summary
CUA-S1-FORMS is a small specialised model from Cua, a San Francisco company, published on 18 September 2026 as the first release in its CUA-S1 family. It does not generate text. Fed the context of a form plus document values already pulled out, it scores each permitted option in one forward pass, so the decision never needs a second call to a large model.
Features
- Single-pass choice: returns a probability for each permitted option in one forward pass.
- Byte-level encoder: a small two-layer transformer reads the form context and each option as bytes.
- Synthetic training: trained on synthetic form episodes with two to sixteen fields, with no form appearing in both training and test sets.
- Local operation: runs on the machine doing the work, with no API call.
- Open release: the weights, training code and synthetic dataset are published under the MIT licence.
Clients
Developers building agents that operate computers and fill in forms from extracted document values.
Key facts
- Parameters: 706,048 trainable parameters, with a checkpoint of 2.8 MB.
- Limits: it can only choose among values already supplied as labelled entities, and it was trained on synthetic data with little real-world validation.
- Language: the encoder is English-focused.
Summary written by TrustList from the business's own published information, Nov 2026.
Benchmarks & AI stats
| Benchmark | Official | Community avg |
|---|---|---|
| No benchmark scores yet. Be the first to add one. | ||
“Official” values are editor-approved and feed the ranking. “Community avg” is the mean of member submissions (shown for transparency; it never affects the ranking until an editor approves a value).
Sign in to add a benchmark score for this model.
Request a demo or quote from CUA-S1-FORMS
Reviews
Write the first review of CUA-S1-FORMS
Used it? Your experience helps other buyers decide.
Write a reviewQuestions & answers
No questions yet. Be the first to ask about CUA-S1-FORMS.
CUA-S1-FORMS alternatives
Similar listings buyers compare against this one.
Cloudflare's multimodal decision model, released 9 October 2026 under Apache 2.0: takes text, images, audio and video and returns a score for each allowed answer instead of free text. 30B mixture of experts with 3B active.
Google's open multimodal embedding model that maps text, images, audio and video into one vector space, under Apache 2.0
Anthropic's smallest and fastest Claude 5.5 model, available on its own API and three cloud platforms
Perplexity's open-weight decision model that picks among given options for routing and checks in AI systems
Aleph Alpha's open-weight English-German mixture-of-experts model for self-hosting, released under Apache 2.0 in October 2026
Cantina Security's first open-weight security model, a post-trained GLM-5.3-Flash for vulnerability research and exploit checks
Microsoft AI's text-to-speech model for keeping one voice across 23 languages, with voice cloning from short audio
More on TrustList
Everything here links back to the same verified catalogue. Pick your next stop.
- Write a reviewHelp other buyers decide
- CompaniesAgencies, consultancies and IT service providers, ranked by verified reviews.
- ProductsSoftware and SaaS with pricing, features, integrations and alternatives.
- ArticlesGuides, comparisons and research from the editorial desk and the community.
- AwardsAnnual recognition decided by verified reviews and an independent jury.
- LaunchesNew products and releases, voted up by the community every day.
- RequestsBuyers describe what they need; vendors respond directly.
- PeopleReviewers, authors and makers with public profiles.
- ComparePut up to four listings side by side before you shortlist.