A 2.8 MB byte-level transformer for GUI form filling. For each form element it returns one probability per typed option (fill an entity, check, click or skip), using jevlike's option-attention head.
The most-downloaded open System One reproduction. Merged Qwen3.5 fine-tunes, trained on about 95 public decision sets, then calibration-aware RL and a rank-64 LoRA, that softmax letter logits at an answer slot.
Decides
choice
choice, score, noul, classify, route
Architecture
cua-s1
decider
Fine-tuned from
—
qwen/qwen3.5-2b-base
License
mit
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
100.0%
80.2%
Calibration error
—
0.038
Valid action rate
—
—
Median latency
—
3.2 ms
p95 latency
—
—
Evaluation suite
196-decision evaluation on real forms
Figures are from each model’s manifest; accuracy and latency are what the publishers report, on their own suites and hardware. Add a third model.
Questions
What is the difference between cua-s1-forms and decider?
cua-s1-forms is from Cua and decider from Mapika. Both have open weights you can download and run. Both answer choice questions. Only decider answers score, noul, classify and route. cua-s1-forms is the smaller model, at 706K parameters to 2.0B. cua-s1-forms is licensed mit; decider, apache-2.0.
Which is more accurate, cua-s1-forms or decider?
They report on different suites — cua-s1-forms 100.0% on 196-decision evaluation on real forms, decider 80.2% on Decider 67-task regression set — so the numbers do not rank them. Test both on your own labelled examples.
Which is cheaper, cua-s1-forms or decider?
cua-s1-forms: Free (open weights). decider: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run cua-s1-forms or decider locally?
Yes, both: systemone pull cua/cua-s1-forms and systemone pull mapika/decider download the weights.
