CUA-S1-FORMS: 706K-parameter System 1 model for form-filling agents
Cua open-sourced CUA-S1-FORMS, a 706K-parameter MIT-licensed decision model that scores form-filling actions in one forward pass instead of calling an LLM per field.
Why it matters
A 2.8 MB checkpoint does in 7-9 ms what a per-field LLM call does in ~270 ms - moving part of the agent loop out of the model and into ordinary code.
What you could build with it
Replace per-field LLM calls in any form-filling agent with CUA-S1-FORMS scoring plus deterministic execution: sub-10 ms decisions, zero token cost per field.
Does it hold up?
Credible for one narrow job - real eval numbers published (99.7% on form tasks) and a working Mac integration exists; unproven outside form-filling.
Built with CUA-S1-FORMS
- Cua open-sources a 706,048-parameter model for filling formsruntimewire · Detailed writeup with eval numbers: 99.7% correct on a form task vs 83.6% for hosted Jev, 7-9 ms local scoring vs 260-280 ms per LLM call.
- FluidInference/FluidUsegithub · Local computer use on Apple silicon: fills forms in real Mac apps via the Accessibility API using CUA-S1-FORMS converted to Core ML, ~1 ms per decision on the Neural Engine, nothing leaves the machine.
Learn more
First spotted on hn: source.
More AI models
ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71DeepSeek V4-Flash official API: public beta with upgraded agent capabilities, Responses API and Codex supportDeepSeek's official V4-Flash is now a public beta API with massively upgraded agent capabilities - benchmark scores…models · JEV 0.69Mistral Forge: enterprise platform for training and continuously improving proprietary modelsMistral AI launched Forge (Sep 28) — a full-lifecycle model training platform (pre-training, SFT, DPO/ODPO, RL…models · JEV 0.69kev: open, trainable Jev-like family of small decision models on Qwen3.5/3.8A family of small decision models (0.8B to 27B) built on Qwen3.5/3.8 that reproduces Jev's prefill-only typed-decision…models · JEV 0.68GPT-Synopsys: OpenAI and Synopsys multi-year deal to build an AI model that operates EDA toolsOpenAI and Synopsys signed a multi-year agreement on Sep 30 to jointly develop GPT-Synopsys, a specialized model that…models · JEV 0.66
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.