Ruoxi Qiu's decision adapter for a frozen Qwen3.8-27B: a 6.3M-parameter LoRA on the last four text layers plus a numerical prior. One shared state feeds isolated question branches; each Choice, Noul or Score question returns probabilities with no generated tokens.
DecisionTune's 395M decision model: ModernBERT-large fine-tuned with a 4 KB scoring head that scores a marker per option, answering a choice or yes/no question in one encoder pass with a probability for every option. Runs locally on CPU or GPU via PyTorch, MLX or ONNX.
Decides
choice, noul, score
choice, noul, route, classify
Architecture
canopy-jev
decisiontune
Fine-tuned from
qwen/qwen3.8-27b
answerdotai/modernbert-large
License
apache-2.0
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
87.4%
—
Calibration error
—
—
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 canopy-jev and decisiontune?
canopy-jev is from Ruoxi Qiu and decisiontune from DecisionTune. Both have open weights you can download and run. Both answer choice and noul questions. Only canopy-jev answers score. Only decisiontune answers route and classify. decisiontune is the smaller model, at 395M parameters to 27B.
Which is more accurate, canopy-jev or decisiontune?
Only canopy-jev publishes an accuracy figure (87.4% on JevBench public set (231 items; 202 correct), the maker's own fresh zero-shot run with the upstream scorer); decisiontune does not, so there is no comparison to make without your own test.
Which is cheaper, canopy-jev or decisiontune?
canopy-jev: Free (open weights). decisiontune: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run canopy-jev or decisiontune locally?
Yes, both: systemone pull camellia86/canopy-jev and systemone pull decision-tune/decisiontune download the weights.
—
—
Evaluation suite
JevBench public set (231 items; 202 correct), the maker's own fresh zero-shot run with the upstream scorer
