A 0.6B replica of the System One idea trained on game environments. A Qwen3-0.6B backbone with an attention-based Choice head that scores a dynamic candidate set for Maze, Snake, ViZDoom and position prediction, shipped with its full training pipeline.
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
choice, score, noul, classify, route
Architecture
nanojev
winnow
Fine-tuned from
qwen/qwen3-0.6b
google/gemma-4-12b-it
License
mit
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
—
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 nanojev and winnow?
nanojev is from TianyuCodings and winnow from EldanRing. Both have open weights you can download and run. Both answer choice, noul and score questions. Only winnow answers classify and route. nanojev is the smaller model, at 600M parameters to 12B. nanojev is licensed mit; winnow, apache-2.0.
Which is more accurate, nanojev or winnow?
Only winnow publishes an accuracy figure (85.7% on JevBench public subset (231 items), Q8_0); nanojev does not, so there is no comparison to make without your own test.
Which is cheaper, nanojev or winnow?
nanojev: Free (open weights). winnow: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run nanojev or winnow locally?
Yes, both: systemone pull tianyu-codings/nanojev and systemone pull eldanring/winnow download the weights.
