Bespoke-Nimble-9B
A Qwen3.5-9B adapter for scoring predefined choices, booleans and rubric levels.
Overview
Bespoke-Nimble-9B is a LoRA adapter that scores permitted answer tokens instead of generating free-form reasoning. The model card's 24 September 2026 update specifies an 8,192-token context limit and up to 255 choices per field. It requires the Qwen3.5-9B base checkpoint. The updated checkpoint uses temperature 1.0 without a separate temperature fit.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Aymorphic analysis
A probability output should be validated on your own labels before it controls routing. Compare error rates at different escalation thresholds, and keep the selected checkpoint and scoring temperature in the evaluation record.
Limitations
Model card reviewed; no calibration study performed. Older checkpoint benchmark and calibration claims were removed.
Key features
- Choices, booleans and rubric scores
- Adapter requires base weights
- Explicit context and choice limits
Use cases
- Bounded document classification
- Scoring defined review criteria
Pricing
Apache-2.0 adapter; account for base-model requirements and inference compute.
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- 18 Sept 2026
- Record updated
- 16 Oct 2026
- Record last checked
- 16 Oct 2026