Jev

Jev is a System One decision model and API from TypeSafe AI that returns typed, calibrated answers such as choices, scores, and booleans in a single parallel pass, aimed at fast structured decisions in software rather than generating text.

Jev is the flagship model from TypeSafe AI and, by the company framing, the first of a new class it calls System One models: models built to make fast, structured decisions rather than write prose. A System One model takes program state as input and returns typed answers and probabilities, one per question, so software can act on the output directly without parsing free text. Technically, Jev departs from the autoregressive pattern of large language models. TypeSafe describes a new architecture with a parallel sampler that generates all outputs in a single query rather than one token at a time, trained with a method it calls Reinforcement Learning for Calibrated Decisions. The company reports a zero percent structured output error rate and end-to-end latency in the range of 70 to 500 milliseconds. Jev currently accepts text input only, evaluating strings, JSON objects, and arrays of text, and it returns typed choices, scores, and booleans with attached probabilities. The launch drew rapid adoption. Coverage reported that Vercel, Cloudflare, LangChain, and Langfuse integrated Jev within days of its September 2026 debut, and that it reached roughly 13 percent of paid Vercel teams within 24 hours. Forbes covered the launch on September 19, 2026, framing Jev as a decision layer that cuts the cost of AI decisions sharply versus prompting a general chatbot. Access is via an API in early access, requiring a waitlisted key, and Jev is also reachable through gateways including OpenRouter, the Vercel AI Gateway, the Netlify AI Gateway, and AIMLAPI. Jev is a fit for teams building software automation that needs reliable typed decisions, and it is not a general-purpose text or chat model.