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.
Uprelic's hosted decision API, run on GPUs in Paris. Answers noul, choice and score questions about text, JSON and up to 8 images, with a probability for every option and no generated text, on the same request shape as Jev. No weights; base model and size not disclosed.
Decides
choice, noul, score
choice, score, noul, classify, route
Architecture
canopy-jev
strom
Fine-tuned from
qwen/qwen3.8-27b
—
License
apache-2.0
proprietary
Availability
Open weights
Hosted API
Hosted by
—
Uprelic
Input price
—
$0.042/MTok
Decision accuracy
87.4%
87.0%
Calibration error
—
0.033
Valid action rate
—
—
Median latency
—
136 ms
p95 latency
—
—
Evaluation suite
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 strom?
canopy-jev is from Ruoxi Qiu and strom from Uprelic. canopy-jev has open weights you can download and run; strom is only available as a hosted API. Both answer choice, noul and score questions. Only strom answers classify and route. canopy-jev is licensed apache-2.0; strom, proprietary.
Which is more accurate, canopy-jev or strom?
They report on different suites — canopy-jev 87.4% on JevBench public set (231 items; 202 correct), the maker's own fresh zero-shot run with the upstream scorer, strom 87.0% on JevBench public set (231 items), the maker's production run of Strom 1.0.7 on 1 Oct 2026 (Open-Jev harness format) — so the numbers do not rank them. Test both on your own labelled examples.
Which is cheaper, canopy-jev or strom?
canopy-jev: Free (open weights). strom: $0.042 / $0 per 1M. Open weights cost nothing per call beyond your own hardware.
Can I run canopy-jev or strom locally?
canopy-jev yes — systemone pull camellia86/canopy-jev downloads its weights. The other is only served as a hosted API.
JevBench public set (231 items; 202 correct), the maker's own fresh zero-shot run with the upstream scorer
JevBench public set (231 items), the maker's production run of Strom 1.0.7 on 1 Oct 2026 (Open-Jev harness format)
