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.
Qwen3.5-4B with a merged rank-16 LoRA distilled from 17,408 teacher questions and 30,000 public training rows, read out as a temperature-scaled softmax over answer-letter logits. Also in 9B, 2B and GGUF.
Decides
choice
choice, score, noul, classify, route
Architecture
cua-s1
jevk5
Fine-tuned from
—
qwen/qwen3.5-4b
License
mit
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
100.0%
78.4%
Calibration error
—
0.035
Valid action rate
—
—
Median latency
—
13.2 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 jevk5?
cua-s1-forms is from Cua and jevk5 from Alibi Serikbay. Both have open weights you can download and run. Both answer choice questions. Only jevk5 answers score, noul, classify and route. cua-s1-forms is the smaller model, at 706K parameters to 4.0B. cua-s1-forms is licensed mit; jevk5, apache-2.0.
Which is more accurate, cua-s1-forms or jevk5?
They report on different suites — cua-s1-forms 100.0% on 196-decision evaluation on real forms, jevk5 78.4% on JevBench public hard tier — so the numbers do not rank them. Test both on your own labelled examples.
Which is cheaper, cua-s1-forms or jevk5?
cua-s1-forms: Free (open weights). jevk5: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run cua-s1-forms or jevk5 locally?
Yes, both: systemone pull cua/cua-s1-forms and systemone pull alibi-serikbay/jevk5 download the weights.
