Business AI
Jev
System One decision model: typed decisions, no text generation
Jev is TypeSafe's decision model. It does not generate text: send app state, a typed question, and your allowed options, and it returns typed decisions with calibrated probabilities - yes/no (Noul), one-of-N (Choice), or a number on your scale (Score). Answers land in 70 to 500ms. Announced September 15, 2026 by Diogo Almeida (InstructGPT co-author); now in beta on OpenRouter at $0.042 per million input tokens with output free. Up to 190x faster and 440x cheaper than LLMs on TypeSafe's published workflows. The weights and training are undisclosed.
At a glance
Primary use case
Route support tickets to the right queue, flag signups that look like fraud, screen messages against content rules, score urgency 1-5, decide if an agent workflow should pause for a human - any judgment software makes millions of times a day where you want a probability, not a paragraph. Beta access via OpenRouter; three question shapes (Noul / Choice / Score), multiple questions per call.
Who's behind it
TypeSafe
- Source
- closed-source
- Platforms
- api
- Model support
- Closed decision model - hosted only, also listed in beta on OpenRouter
- Website
- https://typesafe.ai