Bosun 3.1 1.7B
Hanno Labs' small calibrated decision model. A LoRA on Qwen3-1.7B plus trained decision-token embeddings with stable slots, returning the full distribution over up to 255 caller-defined choices and a null slot for Choice, Score and Noul questions.
Loads with trust_remote_code; serving.json is the authoritative prompt and slot contract. On the maker's 23,900-row DecisionBench, Bosun scores 84.9% with ECE 0.050 (Jev 1.13 in its run: 72.0%); by primitive, 92.2% binary, 84.2% candidate selection and 63.8% ordinal. Hanno Labs notes the model was trained on the task families DecisionBench covers, so this is seen-family, not out-of-domain, evidence. Siblings: Bosun 3.1 0.6B and GGUF builds; earlier Bosun releases predate the System One category. The third-party Decision Index 0.2.1 scores the 1.7B at 20.1.
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 | 2B |
| Base model | qwen/qwen3-1.7b |
| Maker | Hanno Labs |
| Released | 2026-09-22 |
| License | apache-2.0 |
| Reported accuracy | 84.9% |
Get the weights
pip install systemonemodels
systemone pull hanno-labs/bosun
The files are served from the maker's Hugging Face repository, Hanno-Labs/bosun-v3.1-1.7b, and verified against the checksums recorded here.