Kev brings small, local decision models to Qwen 3.5

You don't need a massive cloud LLM just to route support tickets or score user feedback.

Developer Jared Palmer released Kev, an open-source family of small decision models built on top of Qwen3.5. Available in 0.8B, 4B, and 9B sizes, Kev processes yes/no, multiple-choice, and rating questions in a single request. Instead of returning raw text, it gives you probability breakdowns for every option.

Why it matters: Big LLMs often require heavy prompt engineering to output reliable structured data. Kev runs locally on CUDA, ROCm, or Apple Silicon—the 4B and 9B models fit on a 32GB Mac—and isolates each question so earlier answers don't bias later ones.

You can try the 0.8B and 4B models instantly on Hugging Face Spaces without installing anything. If you want to run it locally, the included server hooks directly into TypeSafe's Python SDK using standard uv commands.

Structured decision-making on your own hardware, without the cloud bill.