A training-free layer that turns an open LLM into a decision model. It reads typed choice, yes/no and score answers from one prefill, removes option-order bias with no labels and, from a few hundred labels, calibrates or fits a closed-form head. Weights stay untouched.
The first System One model. Reads a state, answers typed Choice, Score and Noul questions in one call with calibrated probabilities, and generates no text. Closed weights, served by TypeSafe AI.
Decides
choice, score, noul, classify, route
choice, score, noul, classify, route
Architecture
anyjev
jev
Fine-tuned from
qwen/qwen3-8b
—
License
apache-2.0
proprietary
Availability
Open weights
Hosted API
Hosted by
—
TypeSafe AI
Input price
—
$0.042/MTok
Decision accuracy
77.1%
—
Calibration error
0.034
—
Valid action rate
—
—
Median latency
—
—
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 anyjev and jev?
anyjev is from Nokia Applied Research and jev from TypeSafe AI. anyjev has open weights you can download and run; jev is only available as a hosted API. Both answer choice, score, noul, classify and route questions. anyjev is licensed apache-2.0; jev, proprietary.
Which is more accurate, anyjev or jev?
Only anyjev publishes an accuracy figure (77.1% on LocalLLaMA/typed-decisions (2,000 held-out decisions), L2 with 300 labels per question); jev does not, so there is no comparison to make without your own test.
Which is cheaper, anyjev or jev?
anyjev: Free (open weights). jev: $0.042 / $0 per 1M. Open weights cost nothing per call beyond your own hardware.
Can I run anyjev or jev locally?
anyjev yes — systemone pull nokia-applied-research/anyjev downloads its weights. The other is only served as a hosted API.
Evaluation suite
LocalLLaMA/typed-decisions (2,000 held-out decisions), L2 with 300 labels per question
