Bespoke Labs launches Nimble, an open 9B local model for typed text decisions with answer probabilities
Summary
Bespoke Labs launches Nimble, an open 9B-parameter local model that makes typed boolean or enum text decisions with answer probabilities, matching 292 of 324 held-out labels while limiting inputs to 2,048 tokens and schemas to 26 choices.
Key Points
- Bespoke Labs launches Nimble, an open 9B local model and training recipe for making typed text decisions with answer probabilities.
- Bespoke-Nimble-9B matches 292 of 324 synthetic held-out labels (90.12%), outperforming untuned Qwen3.8-27B by 17 matches but trailing TypeSafe's Jev by 10.
- The model is trained on 2,676 contrastively curated examples, supports flat boolean or enum schemas with up to 26 choices, and rejects prompts longer than 2,048 tokens.