Decision Models · checked Oct 17, 2026
AI that decides, not chats.
Decision models answer typed questions (yes/no, pick one, score) with a probability, in one fast pass. They never write prose. Think of them as a smart if-statement for your app or agent.
What is a decision model?
Text, JSON or (from Jev 1.14) an image: a support ticket, a review, a form.
Boolean (“is this a refund request?”), Choice (“which team?”) or Score (“how urgent, 1–5?”).
Every answer comes with a calibrated probability, so you can act when it’s sure and escalate when it isn’t.
question (noul): “Is the customer asking for money back?”
refund: 0.98TypeSafe calls this fast “System One” thinking, next to LLMs as slower “System Two”. Use a decision model to classify, route, score, extract and verify; use an LLM when you need text, code or open-ended reasoning.
Also listed on Vercel
Other decision models Vercel links from its TypeSafe docs page. We list only what that page says; no prices or benchmarks were verified.
Vercel lists it as available on AI Gateway, free through October 31.
Vercel lists it as available on AI Gateway.
Vercel lists it as available on AI Gateway.
Vercel lists the Decider V1.2 models as available on AI Gateway.
Where this fits
Running agents? A decision model is a cheap way to gate tool calls or route tasks inside an agent loop. See open agents and LLM model pages.
FAQ
What is a decision model?
An AI model that answers typed questions (yes/no, choice, score) about some input, with a probability, instead of writing text. Jev from TypeSafe AI is the best-known example.
Is Jev a chatbot?
No. Jev never writes prose. It returns Choice, Score or Boolean answers with calibrated probabilities.
How much does Jev cost?
$0.042 per million input tokens; output is free (TypeSafe docs, checked 2026-10-17).