Decision-1.0-Lux-9B
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.
On the team's 54-task benchmark (3,766 decisions, weighted across general decisions, composition, reading, inference and transfer) Lux-9B scores 77.40%, against 81.05% for Jev and 71.89% for Kev-9B on the same benchmark. Choice takes 2 to 255 options and score 2 to 10 levels. The repository holds the fine-tuned backbone, the candidate head and fitted temperatures; serving runs through vLLM Semantic Router's Decision runtime, which speaks /v1/systemone. Siblings, all Apache-2.0: Nox-4B (73.09%), Sol-2B (66.32%) and Eos-0.8B (61.89%) on Qwen3.5, and Kai-0.6B (53.52%) and Lex-0.6B on the team's Vela encoder, which keep their upstream Gemma tokenizer terms.
What it decides
- choice — picks one option from a set
- score — places the input on an ordered scale
- noul — answers a yes/no question with one calibrated probability
- classify — assigns a category from a fixed taxonomy
- route — sends the input to one of several destinations
At a glance
| Parameters | 9B |
| Base model | qwen/qwen3.5-9b |
| Maker | vLLM Semantic Router |
| Released | 2026-09-22 |
| License | apache-2.0 |
| Reported accuracy | 77.4% |
Get the weights
pip install systemonemodels
systemone pull vllm-semantic-router/decision
The files are served from the maker's Hugging Face repository, llm-semantic-router/Decision-1.0-Lux-9B, and verified against the checksums recorded here.