What is a decision model? How it differs from a chat model

A decision model is an AI model that answers a question about some text with a value your software can use directly, instead of writing a reply. You give it some text and a question, and it gives back one of three kinds of answer: an option picked from a list you wrote, a score on a scale you described, or the probability that a yes/no statement is true. Each answer comes with probabilities.

TypeSafe AI's Jev was the first one most developers met, launched on 15 September 2026. TypeSafe calls this kind of model a System One model. Its documentation describes them as models "built to make fast, structured decisions that software can use directly". The name borrows from Daniel Kahneman's idea of fast, intuitive thinking, as opposed to slow, deliberate thinking.

• What is Jev?

Both understand ordinary language. The difference is what comes back, and what the model is for. According to TypeSafe's documentation:

• It returns a value, not text. The answer has a type your code can check: an option, a number on a scale, or a probability.
• It does not explain itself. TypeSafe says System One models do not write replies, produce code or generate explanations of their reasoning.
• Its probabilities are trained to mean something. TypeSafe says the probabilities are optimised against real outcomes so that they reflect uncertainty. It adds that calibration is measured across groups of predictions and does not guarantee that any one answer is correct.
• It is for quick judgments. TypeSafe's advice is to ask what a knowledgeable person could judge in a second, and to split bigger questions into small ones whose answers you combine in your own code.

• Jev vs ChatGPT, Claude and Gemini

TypeSafe calls the question types primitives, and there are three for text. A Choice picks one option from a fixed set, such as which team should handle a ticket. A Score places something on an ordered scale you describe in words, such as how frustrated a customer is. A Noul answers a yes/no question with the probability that the answer is yes, where 0 means no and 1 means yes.

Choice and Score answers also carry a probability for every option or level and a confidence value. A Noul answer is a single probability. You can send several questions about the same text in one request, and each comes back under the name you gave it.

• How to use Jev

The jobs TypeSafe's documentation uses as examples are all small judgments inside a larger piece of software:

• Routing: which team should handle this support ticket.
• Flagging: does this message ask for a refund, or contain personal data.
• Rating: how severe is this bug, or how frustrated is this customer.
• Gates: whether the evidence supports an action, so that your code decides whether to proceed or send the case to a person.

• Use cases for Jev, with the cost of each

Anything that needs words back. A decision model will not write an email, explain why it chose an answer, or summarise a document. Jev reads text only by default, and TypeSafe says its image input is experimental. TypeSafe's own guidance is that a question needing slow reasoning is a sign to break it into small questions, or to pass it to a reasoning model.

Its probabilities also have limits. They are meant to be honest on average across many answers, not a promise that one answer is right, so it is worth testing a decision model on a labelled sample of your own data before you rely on it.

• What Jev can't do

Jev came first, and the category has grown quickly since. OpenAI's Decisions API, which uses the model gpt-6-luna, became generally available on 22 October 2026, after a public beta from 6 October. Perplexity, Microsoft, Liquid AI and Snowflake have each released a model for the same job, and several are open: AWS, Cloudflare and Convai among them. Each is compared with Jev, with sources, in the articles below.

Prices differ. Jev lists $0.042 per million input tokens with output free, and OpenAI lists $0.10 per million uncached input tokens, also with output free. That is each company's list price on the dates we checked, not a measured cost for a whole task.

• Jev vs OpenAI's Decisions API
• Jev vs Perplexity Decider
• Jev vs Microsoft-Decision-1
• Jev vs open decision models

The interesting change is not that these models are smarter than chat models. It is the shape of the answer. A value with a probability is something your code can test, log and compare, which a paragraph of text is not. That makes a decision model a part you can slot into software, and it is why the idea spread across companies within weeks.

It is also a young category, under two months old in public, and most of the speed, accuracy and cost figures around it are the vendors' own. The safe way to use one is to write a few clear questions, run a labelled sample of your own data through it, and look at the answers before you trust them.

• Jev pricing and what it costs