Sam Witteveen · 2026-09-18 · notable
Sam Witteveen — 'Jev - The Ultimate Classification Model?'
Sam Witteveen's 18 September 2026 video asks whether Jev is the ultimate classification model. Jev is TypeSafe AI's first System One Model: it answers typed questions with typed values and a confidence score, at $0.042 per 1M input tokens.
Sam Witteveen's channel turns to Jev — the model that answers typed questions instead of writing text.
What is it?
'Jev - The Ultimate Classification Model?' went up on Sam Witteveen's channel on 18 September 2026. Its subject is Jev, the first System One Model from TypeSafe AI, which opened early access on 15 September. Jev takes unstructured text standing in for program state plus typed questions, and returns type-safe values with a calibrated probability instead of prose.
How does it work?
Jev samples in parallel rather than generating tokens one after another, which TypeSafe puts at 70–500 milliseconds per call. Every answer carries a calibrated confidence score, and TypeSafe states the model never makes type errors, so calling code can branch on a result without parsing it first. Input costs $0.042 per million tokens and output tokens are free.
Why does it matter?
Classification and routing are what a large share of production LLM calls actually do, and they are usually served by a general chat model that bills for every output token. The question Witteveen's title raises is therefore a cost question as much as a quality one: TypeSafe charges nothing for output and quotes sub-second responses, against frontier models it says are 40–200x slower for comparable intelligence. Jev is still early access, with developers added from a waitlist.
Who is it for?
engineers using LLMs for classification and routing