A multimodal decision classifier over text, image, audio and video on Gemma 4 12B, fine-tuned on 30,000 questions. Returns a probability per option and generates nothing.
A Gemma 4 12B fine-tune for typed decisions that also chats and reads images. Ships as GGUF for llama.cpp, holds a 64K context on a 16 GB GPU, and serves /v1/systemone next to /v1/chat/completions.
Decides
choice, noul, score, classify
choice, score, noul, classify, route
Architecture
jev-omni
winnow
Fine-tuned from
google/gemma-4-12b-it
google/gemma-4-12b-it
License
apache-2.0
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
87.6%
85.7%
Calibration error
0.040
—
Valid action rate
—
—
Median latency
83 ms
143 ms
p95 latency
—
—
Evaluation suite
DecisionBench Medium (author's set)
JevBench public subset (231 items), Q8_0
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 jev-omni and winnow?
jev-omni is from akhilaaa3 and winnow from EldanRing. Both have open weights you can download and run. Both answer choice, noul, score and classify questions. Only winnow answers route.
Which is more accurate, jev-omni or winnow?
They report on different suites — jev-omni 87.6% on DecisionBench Medium (author's set), winnow 85.7% on JevBench public subset (231 items), Q8_0 — so the numbers do not rank them. Test both on your own labelled examples.
Which is cheaper, jev-omni or winnow?
jev-omni: Free (open weights). winnow: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Which is faster, jev-omni or winnow?
By their publishers’ figures, jev-omni answers in about 83 ms at the median and winnow in about 143 ms — measured on different hardware, so treat it as a rough guide.
Can I run jev-omni or winnow locally?
Yes, both: systemone pull akhilaaa3/jev-omni and systemone pull eldanring/winnow download the weights.