TokenRhythm's prefill-only decision model for agent workflows. Built on its NeoHorse-1-4B (a Qwen3.5-4B derivative), it predicts Choice, Noul and Score answers over application-defined options without generating text, with optional single-image input.
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, score, noul, classify, route
choice, score, noul, classify, route
Architecture
neohorse
winnow
Fine-tuned from
tokenrhythm/neohorse-1-4b
google/gemma-4-12b-it
License
apache-2.0
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
75.3%
85.7%
Calibration error
—
—
Valid action rate
—
—
Median latency
—
143 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 neohorse-jev and winnow?
neohorse-jev is from TokenRhythm and winnow from EldanRing. Both have open weights you can download and run. Both answer choice, score, noul, classify and route questions. neohorse-jev is the smaller model, at 4.0B parameters to 12B.
Which is more accurate, neohorse-jev or winnow?
They report on different suites — neohorse-jev 75.3% on JevBench public set (231 items), vLLM, maker's run, 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, neohorse-jev or winnow?
neohorse-jev: Free (open weights). winnow: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run neohorse-jev or winnow locally?
Yes, both: systemone pull tokenrhythm/neohorse-jev and systemone pull eldanring/winnow download the weights.
Evaluation suite
JevBench public set (231 items), vLLM, maker's run
