Kev ships 0.8B/4B/9B open-weight decision models that drop into TypeSafe's Jev API, 1,684 stars in four days
Jared Palmer published Kev (github.com/jaredpalmer/kev) on 2026-09-17; it reached 1,684 stars by 09-21. It is a family of small decision models built on Qwen3.5 following the architecture reverse-engineered in "Jev's Architecture Unmasked," released Apache-2.0 with training code, frozen eval suites on HuggingFace and pretrained 0.8B, 4B and 9B weights. One request carries yes/no (noul), multiple-choice (choice) and rating (score) questions that share the input text but cannot read each other, and the API matches TypeSafe's System One so their Python SDK can be pointed at a local server. The 4B and 9B fit a 32 GB Mac in bf16, which makes the closed Jev classifier service locally replaceable for routing, triage and guardrail work.
Source
↳ Follow the thread