An open Jev-style LoRA on Qwen3.5-9B from Bespoke Labs, trained on 2,676 contrastively curated examples to score the allowed answer tokens directly for enums, booleans and rubric levels. Recipe, data and a public benchmark suite are released with it.
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, route
choice, score, noul, classify, route
Architecture
nimble
winnow
Fine-tuned from
qwen/qwen3.5-9b
google/gemma-4-12b-it
License
apache-2.0
apache-2.0
Availability
Open weights + hosted API
Open weights
Hosted by
Bespoke Labs
—
Input price
—
—
Decision accuracy
90.1%
85.7%
Calibration error
0.054
—
Valid action rate
—
—
Median latency
106 ms
143 ms
p95 latency
—
—
Evaluation suite
Bespoke held-out set (324 examples)
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 bespoke-nimble-9b and winnow?
bespoke-nimble-9b is from Bespoke Labs and winnow from EldanRing. bespoke-nimble-9b has open weights and a hosted API; winnow has open weights you can download and run. Both answer choice, noul, score, classify and route questions. winnow reads up to 64K tokens of state, against 8K tokens for bespoke-nimble-9b. bespoke-nimble-9b is the smaller model, at 9.0B parameters to 12B.
Which is more accurate, bespoke-nimble-9b or winnow?
They report on different suites — bespoke-nimble-9b 90.1% on Bespoke held-out set (324 examples), 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, bespoke-nimble-9b or winnow?
bespoke-nimble-9b: Hosted, price not published, or free to self-host. winnow: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Which is faster, bespoke-nimble-9b or winnow?
By their publishers’ figures, bespoke-nimble-9b answers in about 106 ms at the median and winnow in about 143 ms — measured on different hardware, so treat it as a rough guide.
Can I run bespoke-nimble-9b or winnow locally?
Yes, both: systemone pull bespoke-labs/bespoke-nimble-9b and systemone pull eldanring/winnow download the weights.
