The flagship of the vLLM Semantic Router team's Decision 1.0 family. Qwen3.5-9B with a shared candidate head returns a probability for every supplied answer to choice, yes/no and score questions, over a 16,384-token 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
decision
winnow
Fine-tuned from
qwen/qwen3.5-9b
google/gemma-4-12b-it
License
apache-2.0
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
77.4%
85.7%
Calibration error
—
—
Valid action rate
—
—
Median latency
—
143 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 decision and winnow?
decision is from vLLM Semantic Router and winnow from EldanRing. Both have open weights you can download and run. Both answer choice, score, noul, classify and route questions. winnow reads up to 64K tokens of state, against 16K tokens for decision. decision is the smaller model, at 9.0B parameters to 12B.
Which is more accurate, decision or winnow?
They report on different suites — decision 77.4% on vLLM-SR decision benchmark (54 tasks, 3,766 decisions, weighted), 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, decision or winnow?
decision: Free (open weights). winnow: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run decision or winnow locally?
Yes, both: systemone pull vllm-semantic-router/decision and systemone pull eldanring/winnow download the weights.
