> ## Content Index
> Fetch the complete content index at: https://globalfeed.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# TypeSafe AI launches Jev, a model that decides instead of writing text
- URL: https://globalfeed.ai/en/typesafe-ai-launches-jev-a-model-that-decides-instead-of-writing-text/
- Published: 2026-09-18T21:28:03.000Z
- Updated: 2026-09-18T21:28:03.000Z
- Description: TypeSafe AI came out of two years of stealth to unveil Jev, its first model. Unlike LLMs, this 'System One' model generates no text and returns type-safe decisions. It is named after the Jevons paradox, and its founder helped build the methods behind ChatGPT.
- Author: GlobalFeed Editor
- Tags: TypeSafe AI, Jev, Diogo Almeida, System One, x-typesafeai, dil-en, elle, video

AI lab [TypeSafe AI](https://x.com/typesafeai?ref=globalfeed.ai) emerged from two years of stealth on 16 September to unveil its first model, Jev. The company's founder and chief executive is Diogo Almeida ([@CompleteSkeptic](https://x.com/CompleteSkeptic?ref=globalfeed.ai)), who helped build the RLHF and InstructGPT methods behind ChatGPT. TypeSafe raised a $40 million seed round led by DCVC.

## What Jev does

Jev is the first of what the company calls System One models. Its defining trait: it generates no text. It takes an unstructured state plus a predefined set of options and returns a type-safe decision: a yes/no probability, a choice from a list, or a score on a scale, each with a calibrated probability and confidence score. In the company's words, it is a function call for frontier intelligence: unstructured state in, typed probabilistic decisions out.

Because the output is constrained to a predefined schema, TypeSafe says the model cannot hallucinate and never makes type errors. It calls the training method RLCD, reinforcement learning for calibrated decisions.

**A caveat:** the no-hallucination guarantee only covers the structure of the output. Picking the wrong option among the allowed ones is still possible, so you have to validate on your own data that the model decides correctly.

## How it differs from other models

The real break is in how it produces an answer. Today's language models generate text token by token, sequentially: slow, costly, and able to invent a source that does not exist. Jev instead returns a decision in a single, parallel pass; it cannot produce a string that does not exist, because the output is always a valid member of the schema.

Almeida frames it with a historical analogy: replacing sequential computation with parallel is like Transformers leaving RNNs behind. It is also a different path from System 2 reasoning models that think slowly: Jev exists to make the millions of small decisions inside a workflow, routing, classifying, judging, choosing, quickly and cheaply, not for open-ended generation.

## Price, speed and demos

The pricing claim stands out: $0.042 per million input tokens ($42 per billion, in Almeida's phrasing), with output tokens free, too cheap to meter. The company reports end-to-end response times of 70-500 milliseconds on System One-shaped queries, against 3-329 seconds for frontier LLMs. Two demos make the case: Jev plays Doom in real time (about 10 queries per second, roughly $7 an hour) and runs a Wikirace, moving from one Wikipedia page to another using only in-page links, to show the compounding benefit of not hallucinating when choosing among hundreds to thousands of options.

**An important caveat:** these speed and cost figures are not independently verified. TypeSafe built the comparison workflows itself and used other models' outputs as the reference answers, and some strong models were left out. The company itself concedes possible team bias, and the model's architecture is undisclosed.

## Why Jev

The name comes from the 19th-century economist William Stanley Jevons and his paradox: efficiency gains raise resource use rather than cutting it. That is TypeSafe's bet, that as intelligence gets cheaper, usage explodes. Almeida calls Jev the shortest path to an AI-based economic revolution.

Sources: [TypeSafe AI blog](https://typesafe.ai/blog/introducing-system-one-models-and-jev?ref=globalfeed.ai), [SiliconANGLE](https://siliconangle.com/2026/09/16/typesafe-ai-exits-stealth-with-40m-to-build-ai-for-use-by-software/?ref=globalfeed.ai).