← Index

ROUTEAI infra

Jev: An Independent Index for Decision Making

Alex Freitas@AlexFreitasAI
Project photo 1

Jev is a model that takes program state and typed questions as input, returning calibrated yes/no answers, choices, or scores with low latency, designed to handle decision-making tasks separate from text generation.

TakeIt returns calibrated yes/no, choices, or scores in 70 to 500ms.

Pattern↑ Jev AI decision models

ChatGPT's RLHF co-inventor just shipped a model that refuses to chat. Diogo Almeida (TypeSafe) launched Jev on Sep 15. No text generation. You hand it program state plus typed questions. It returns calibrated yes/no, choices, or scores in 70 to 500ms. Input is $0.042 per million tokens. Output is free. Keep the LLM for writing. Offload the judgment calls: route, escalate, score, jailbreak check. That's how you cut latency and bill without dropping intelligence on the branch. Early access at typesafe.ai, or Vercel AI Gateway as typesafe-ai/jev. Which decision in your stack still burns a full frontier call?
Original post