What it is
Jev is TypeSafe's flagship model and the first "System One" model. You send it a state, which is your data, plus typed questions about that state. It returns structured answers your code can use directly, without generating or parsing text1. Answers come in typed forms such as choices, scores, truth values, and image bounding boxes, along with probabilities1.
My plan
Jev is billed by usage, so there's no monthly plan price. As of November 5, 2026, TypeSafe's home page lists "$42" per billion input tokens2. The Register's launch coverage gives the same rate as $0.042 per million input tokens and $0 for output3. TypeSafe lists no plans or tiers on its home page, and its docs introduction doesn't mention pricing2,1. Jev is in early access2.
What it's good at
TypeSafe's docs list three strengths: typed outputs that code can branch on, sort by, and route with; independent evaluation of each question, so adding questions doesn't create context rot; and calibrated probabilities, with confidence on some question types1. TypeSafe claims a response time of 70ms to 500ms, which The Register reports as company claims3.
The limits are stated plainly in the docs. Jev is designed for well-scoped, atomic questions, and questions that need extended reasoning or weigh several independent factors should be split up, with the results combined in your own code1. Confidence reporting is available only for some question types1. The Register also notes that TypeSafe's "hallucination-free" claim compares Jev to chat models unfairly, because Jev's output isn't natural language3.
What I use it for
Typed AI judgments inside the software I build.
The company
TypeSafe AI describes itself as "an AI lab building machine-native intelligence infrastructure for automation"2. Diogo Almeida is its co-founder and CEO, a former OpenAI researcher3. The Register reported Jev's launch on September 16, 2026, and described TypeSafe as a startup with $40 million in funding3. I couldn't confirm TypeSafe's founding year or headquarters from a page I could open, so I'm leaving them out.
Sign up
typesafe.ai, with docs at docs.typesafe.ai
Sources
- Introduction. TypeSafe AI (docs.typesafe.ai), dated November 5, 2026.DocumentationRead
What it says: Jev is TypeSafe's flagship model and the first System One model: you send a state plus typed questions and get structured typed answers with probabilities, no text to parse. Strengths: typed outputs, independent evaluation of questions, calibrated probabilities. Limits: built for well-scoped atomic questions; confidence only for some question types. No pricing on the page; undated, checked today.
- Home - TypeSafe AI. TypeSafe AI, checked November 5, 2026.Vendor pageRead
What it says: TypeSafe is "an AI lab building machine-native intelligence infrastructure for automation"; Jev is in early access. Lists "$42" per billion input tokens. No plans or tiers listed; founders and headquarters not stated.
- TypeSafe AI debuts model for machines that plays Doom. The Register, dated September 16, 2026.ArticleRead
What it says: Jev returns typed probabilistic outputs; claimed 70ms-500ms response time; pricing $0.042/MTok input and $0 output. TypeSafe has $40 million in funding; Diogo Almeida, a former OpenAI researcher, is co-founder and CEO. Notes TypeSafe's hallucination-free claim is an unfair comparison. Does not state headquarters.