Typed decisions from Google's frozen Gemma-4-12B-it, with no fine-tuning. A small shim over unmodified vLLM letters the options, reads the model's own probability for each letter at one answer position and applies one calibration temperature; one output token per decision.
An independent Jev-shaped reproduction on DeBERTa-v3-large. One state and any number of choice, score and yes/no questions go in, and a calibrated distribution per question comes out of one forward pass. Public gold labels only.
Decides
choice, score, noul
choice, score, noul, classify
Architecture
cygnet
open-jev
Fine-tuned from
google/gemma-4-12b-it
microsoft/deberta-v3-large
License
mit
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
87.9%
85.4%
Calibration error
—
0.022
Valid action rate
—
—
Median latency
50 ms
28 ms
p95 latency
—
—
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 cygnet and open-jev-deberta-v3-large?
cygnet is from Blockbrain Labs and open-jev-deberta-v3-large from Kotoba Labs. Both have open weights you can download and run. Both answer choice, score and noul questions. Only open-jev-deberta-v3-large answers classify. cygnet reads up to 16K tokens of state, against 512 tokens for open-jev-deberta-v3-large. open-jev-deberta-v3-large is the smaller model, at 434M parameters to 12B. cygnet is licensed mit; open-jev-deberta-v3-large, apache-2.0.
Which is more accurate, cygnet or open-jev-deberta-v3-large?
They report on different suites — cygnet 87.9% on JevBench public set (231 items), JevBench CLI, open-jev-deberta-v3-large 85.4% on Kotoba held-out test, seen question types (banking77, SST-5, BoolQ) — so the numbers do not rank them. Test both on your own labelled examples.
Which is cheaper, cygnet or open-jev-deberta-v3-large?
cygnet: Free (open weights). open-jev-deberta-v3-large: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Which is faster, cygnet or open-jev-deberta-v3-large?
By their publishers’ figures, open-jev-deberta-v3-large answers in about 28 ms at the median and cygnet in about 50 ms — measured on different hardware, so treat it as a rough guide.
Can I run cygnet or open-jev-deberta-v3-large locally?
Yes, both: systemone pull blockbrain-labs/cygnet and systemone pull kotoba-labs/open-jev-deberta-v3-large download the weights.
Evaluation suite
JevBench public set (231 items), JevBench CLI
Kotoba held-out test, seen question types (banking77, SST-5, BoolQ)
