Jev decision model lands on Vercel AI Gateway at $0.04 per million tokens

TypeSafe AI's Jev is built to answer questions for software rather than to chat. According to its listing on Vercel's AI Gateway, a developer sends Jev some shared state, meaning text that describes a situation, together with typed questions about it, and Jev returns choices, scores, or probabilities for yes-or-no questions. The listing's code example passes the sentence "The support agent issued a full refund to the customer" with one boolean question, "Was a refund issued?", through an experimental decide function in Vercel's AI SDK. The application gets back a value it can test in an if statement instead of a paragraph it would have to parse. Several questions about the same state can be evaluated in parallel within one request.

TypeSafe calls Jev a "System One" decision model, a term psychology uses for fast, intuitive judgment, and the listing names classification, routing, rubric-based assessment and automated verification as its uses. Those are the points in a program where a yes, a no, or a category decides what happens next, such as sending a support ticket to the right queue or checking whether an agent's action matched a policy. The listing prices Jev from $0.04 per million input tokens with a 32K-token context, served by TypeSafe AI and DigitalOcean, and shows versions dated September 15 and October 15, 2026. It gives no accuracy or speed results, so the listing establishes what Jev returns and what it costs, not how often its answers are right. TypeSafe's own launch announcement followed a few days after the gateway post.