Jev is an AI model from TypeSafe.ai, and the first in a category the company calls System One. You send it a piece of text and a set of typed questions, and it returns a structured answer for each one with probabilities attached. It does not write prose, so there is nothing for your code to parse. The version we tested was jev-1.13.0, and the listed price at the time of writing was $0.042 per million input tokens with no charge for output.

What is a System One model?

TypeSafe describes System One models as models that make fast, structured decisions that software can use directly. The name comes from Daniel Kahneman's book Thinking, Fast and Slow, in which System 1 is fast, intuitive thinking and System 2 is slow, deliberate reasoning. A System One model gives a quick judgment on a narrow question. It does not reason through multi-step problems, and it does not write replies, code or explanations.

What are Noul, Choice and Score questions?

They are the three question types Jev answers. A Noul asks whether a statement is true and returns one number from 0 to 1, the probability that the answer is yes. A Choice asks which option from a fixed list applies and returns the chosen option, a probability for every option and a confidence figure. A Score asks which level on an ordered, described scale applies and returns a position on the scale, a probability for every level and a confidence figure. A Noul near 0.5 means the model is unsure. It does not mean a medium amount, so use a Score to measure how much of something there is.

Is Jev a large language model like ChatGPT or Claude?

Jev works differently from a chat model. A chat model writes text, and your code then has to read that text and decide what it means. Jev writes no prose. It returns a probability for each typed question you ask, so the output is already a number or a label your code can compare with a threshold. TypeSafe's documentation says jev-1.13 is not trained to generate text, which fits its purpose: fast, narrow decisions over text, with the reasoning and arithmetic left to your own software.

How is Jev different from a reasoning model?

A reasoning model works through a problem step by step before it answers, which makes it good at maths, code and planning and makes each answer slower and more expensive. Jev does the opposite job. In our tests it answered up to 11 questions about a piece of text in about 0.7 seconds, and TypeSafe's documentation says it handles multi-step reasoning, arithmetic, counting and date comparison poorly. The two are complementary: Jev can sort and screen text quickly, and pass the hard cases to a reasoning model or a person.

How do you call the Jev API?

Send an HTTP POST to https://api.typesafe.ai/v1/systemone with an Authorization bearer key. The JSON body has three fields: model (we pin jev-1.13.0), state (the text to read) and questions (a map of question ids to a type of noul, choice or score, plus instructions and, for choice and score, the options or levels). The response holds one typed answer per question id and a usage block with the input token count. TypeSafe publishes Python and JavaScript SDKs that wrap the call.

How much does Jev cost?

At the time of writing TypeSafe listed Jev at $0.042 per million input tokens, with output free. In our tests a request with three questions about a short email used 471 input tokens, and a request with 11 questions about a news headline used about 1,120. Scoring our archive of 1,499 news headlines (1,508 before duplicates were removed) with 11 questions each cost us $0.0686 in total.

What does confidence mean in a Jev answer?

Every Choice and Score answer includes a probability for each option or level and a single confidence figure from 0 to 1 derived from the shape of that distribution. If nearly all the probability sits on one option, confidence is high. If it is spread across several, confidence is low. In one of our tests Jev split 0.54 to 0.46 between two teams and reported confidence of 0.38. TypeSafe suggests acting automatically at high confidence, proceeding with caution at medium confidence, and sending low-confidence cases to a person or another system. High confidence describes the shape of the distribution. Jev can still be wrong at high confidence, so check it against examples a person has labelled before you let it act.

TypeSafe's own documentation lists the limits for jev-1.13. It does not count reliably, it is poor at arithmetic, it reads dates as text so ordering them and working out durations is unreliable, it gets less accurate when the input contains a lot of irrelevant text, it struggles with double negatives and multi-step reasoning, it takes instructions literally, text written to steer it can move its answers, and it is not trained to generate text. It reads text only and works best in English. The practical rule is to let Jev judge short pieces of text and to do every count, sum, date calculation and threshold in your own code.

What is Jev good for in a business?

Narrow decisions over text where the answer is one of a fixed set: routing a support ticket to a team, deciding whether an email is a complaint or a refund request, rating how urgent a message is, checking whether a supplier message reports a delay, or screening text before it reaches a larger AI model. It suits jobs that are currently done by keyword rules, because it can tell a problem from the same problem being fixed, and it reports how sure it is so that unclear cases can go to a person.

What Is Jev? TypeSafe.ai's System One Model, Explained with Live Examples

Jev is TypeSafe.ai's first System One AI model. It makes fast, structured decisions about text and returns typed answers with probabilities instead of prose. Cloud First Consulting tested jev-1.13.0 independently against the live API on 19 September 2026, and every example, latency, token count and cost in this guide comes from our own calls. You send it a piece of text and a set of typed questions and it returns a probability for each answer instead of prose, at a listed $0.042 per million input tokens with no charge for output. There are three question types. A Noul asks whether a statement is true and returns one number from 0 to 1. A Choice picks from a fixed list of options and returns a probability for every option and a confidence figure. A Score places the text on an ordered scale of described levels. TypeSafe names the category after Daniel Kahneman's System 1, the fast and intuitive half of Thinking, Fast and Slow, and says System One models make fast, structured decisions that software can use directly and do not write replies, code or explanations. This guide records examples run against the live API on 19 September 2026 with jev-1.13.0 pinned. One customer email about a double charge came back with a refund probability of 0.85, a department of billing at confidence 1.00 and a frustration score of 2.98 out of 3, in 471 input tokens and 0.68 seconds. A vague message split 0.54 to 0.46 between two teams and reported confidence of 0.38, which is how Jev reports that it is unsure. An outage headline scored 0.97 for a problem happening now, and the same outage reported as fixed scored 0.98 for an earlier problem resolved, a distinction keyword rules cannot make. The guide also covers the API call (an HTTP POST to https://api.typesafe.ai/v1/systemone carrying model, state and questions), TypeSafe's design rules for building with System One, the published limits of jev-1.13 on counting, arithmetic, dates, long input, double negatives and text generation, the cost we measured (471 input tokens for a three-question email, about 1,120 for an 11-question headline), how Jev compares with chat models such as ChatGPT and Claude, with reasoning models and with keyword rules, and when to use Jev, a chat model or a plain rule. We tested Jev in BTCBuzz, an educational Bitcoin signal bot that paper trades only with no real money, where it read 1,499 archived headlines beside a keyword scorer across 1,451 calls for $0.0686 with no failures, and that case study is a companion article.

Frequently Asked Questions

What is Jev?: Jev is an AI model from TypeSafe.ai, and the first in a category the company calls System One. You send it a piece of text and a set of typed questions, and it returns a structured answer for each one with probabilities attached. It does not write prose, so there is nothing for your code to parse. The version we tested was jev-1.13.0, and the listed price at the time of writing was $0.042 per million input tokens with no charge for output.
What is a System One model?: TypeSafe describes System One models as models that make fast, structured decisions that software can use directly. The name comes from Daniel Kahneman's book Thinking, Fast and Slow, in which System 1 is fast, intuitive thinking and System 2 is slow, deliberate reasoning. A System One model gives a quick judgment on a narrow question. It does not reason through multi-step problems, and it does not write replies, code or explanations.
What are Noul, Choice and Score questions?: They are the three question types Jev answers. A Noul asks whether a statement is true and returns one number from 0 to 1, the probability that the answer is yes. A Choice asks which option from a fixed list applies and returns the chosen option, a probability for every option and a confidence figure. A Score asks which level on an ordered, described scale applies and returns a position on the scale, a probability for every level and a confidence figure. A Noul near 0.5 means the model is unsure. It does not mean a medium amount, so use a Score to measure how much of something there is.
Is Jev a large language model like ChatGPT or Claude?: Jev works differently from a chat model. A chat model writes text, and your code then has to read that text and decide what it means. Jev writes no prose. It returns a probability for each typed question you ask, so the output is already a number or a label your code can compare with a threshold. TypeSafe's documentation says jev-1.13 is not trained to generate text, which fits its purpose: fast, narrow decisions over text, with the reasoning and arithmetic left to your own software.
How is Jev different from a reasoning model?: A reasoning model works through a problem step by step before it answers, which makes it good at maths, code and planning and makes each answer slower and more expensive. Jev does the opposite job. In our tests it answered up to 11 questions about a piece of text in about 0.7 seconds, and TypeSafe's documentation says it handles multi-step reasoning, arithmetic, counting and date comparison poorly. The two are complementary: Jev can sort and screen text quickly, and pass the hard cases to a reasoning model or a person.
How do you call the Jev API?: Send an HTTP POST to https://api.typesafe.ai/v1/systemone with an Authorization bearer key. The JSON body has three fields: model (we pin jev-1.13.0), state (the text to read) and questions (a map of question ids to a type of noul, choice or score, plus instructions and, for choice and score, the options or levels). The response holds one typed answer per question id and a usage block with the input token count. TypeSafe publishes Python and JavaScript SDKs that wrap the call.
How much does Jev cost?: At the time of writing TypeSafe listed Jev at $0.042 per million input tokens, with output free. In our tests a request with three questions about a short email used 471 input tokens, and a request with 11 questions about a news headline used about 1,120. Scoring our archive of 1,499 news headlines (1,508 before duplicates were removed) with 11 questions each cost us $0.0686 in total.
What does confidence mean in a Jev answer?: Every Choice and Score answer includes a probability for each option or level and a single confidence figure from 0 to 1 derived from the shape of that distribution. If nearly all the probability sits on one option, confidence is high. If it is spread across several, confidence is low. In one of our tests Jev split 0.54 to 0.46 between two teams and reported confidence of 0.38. TypeSafe suggests acting automatically at high confidence, proceeding with caution at medium confidence, and sending low-confidence cases to a person or another system. High confidence describes the shape of the distribution. Jev can still be wrong at high confidence, so check it against examples a person has labelled before you let it act.
What can Jev not do?: TypeSafe's own documentation lists the limits for jev-1.13. It does not count reliably, it is poor at arithmetic, it reads dates as text so ordering them and working out durations is unreliable, it gets less accurate when the input contains a lot of irrelevant text, it struggles with double negatives and multi-step reasoning, it takes instructions literally, text written to steer it can move its answers, and it is not trained to generate text. It reads text only and works best in English. The practical rule is to let Jev judge short pieces of text and to do every count, sum, date calculation and threshold in your own code.
What is Jev good for in a business?: Narrow decisions over text where the answer is one of a fixed set: routing a support ticket to a team, deciding whether an email is a complaint or a refund request, rating how urgent a message is, checking whether a supplier message reports a delay, or screening text before it reaches a larger AI model. It suits jobs that are currently done by keyword rules, because it can tell a problem from the same problem being fixed, and it reports how sure it is so that unclear cases can go to a person.

Our Services

Contact Cloud First Consulting

Email: info@cloudfirstconsulting.com

Location: London, United Kingdom

Hours: Monday-Friday, 9:00 AM - 6:00 PM GMT

Book a Free 30-min Discovery Call