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.
Omar (kouhxp)'s CPU decision model: a Qwen3.5-0.8B fine-tune shipped as GGUF for llama.cpp that answers yes/no, choice and score questions with a probability per option plus a 'none of these' reject probability, served by a local Jev-style HTTP runtime.
Decides
choice, score, noul, classify, route
choice, score, noul, classify, route
Architecture
anyjev
gutsy
Fine-tuned from
qwen/qwen3-8b
qwen/qwen3.5-0.8b
License
apache-2.0
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
77.1%
73.2%
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 gutsy?
anyjev is from Nokia and gutsy from Omar (kouhxp). Both have open weights you can download and run. Both answer choice, score, noul, classify and route questions. gutsy is the smaller model, at 800M parameters to 8.0B.
Which is more accurate, anyjev or gutsy?
They report on different suites — anyjev 77.1% on LocalLLaMA/typed-decisions (2,000 held-out decisions), L2 with 300 labels per question, gutsy 73.2% on JevBench public set (231 items; 169 correct), the maker's own run with Q8_0 — so the numbers do not rank them. Test both on your own labelled examples.
Which is cheaper, anyjev or gutsy?
anyjev: Free (open weights). gutsy: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run anyjev or gutsy locally?
Yes, both: systemone pull nokia/anyjev and systemone pull kouhxp/gutsy download the weights.
Evaluation suite
LocalLLaMA/typed-decisions (2,000 held-out decisions), L2 with 300 labels per question
JevBench public set (231 items; 169 correct), the maker's own run with Q8_0
