Jared Palmer Releases Kev, Open Qwen3.5-Based Decision Models for Local Use and Fine-Tuning
Summary
Jared Palmer releases Kev, an open Qwen3.5-based family of 0.8B, 4B and 9B decision models for local deployment and fine-tuning, with calibrated Kev-9B cutting confident errors on new sources from 8.7% to 4.0% while retaining accuracy.
Key Points
- Jared Palmer releases Kev, an open family of 0.8B, 4B and 9B Qwen3.5-based decision models that users can run locally or fine-tune.
- A September 21 second training pass lifts Kev-9B's new-source test accuracy from 0.837 to 0.852, while Kev-4B rises to 0.837 and Kev-0.8B to 0.684.
- Kev serves TypeSafe System One-compatible yes/no, choice and rating requests, and its calibrated Kev-9B reduces new-source confident errors from 8.7% to 4.0% without changing accuracy.