What really is JEV and how does it work?
A new kind of AI that doesn't talk back. It just picks.
You have probably heard of ChatGPT by now. Maybe someone you know showed it to you, or you saw it on the news. You type a question, and it writes back a whole answer, like a very fast pen pal.
In September 2026, a small company in San Francisco called TypeSafe AI released something different. It's called Jev (often written JEV), and it can't write a single sentence. That sounds like a step backwards. It isn't.
The man behind it, Diogo Almeida, used to work at OpenAI, where he helped develop the methods that made ChatGPT good at talking with people. He left to build an AI that talks to computers instead. His company came out of hiding on September 15, 2026, with Jev and $40 million in funding.
A writer versus a picker
The easiest way to understand Jev is to think of two kinds of helpers.
ChatGPT is a writer. Ask it whether an email is urgent, and it writes you a little paragraph: "Yes, this looks urgent because the customer mentions a deadline..." It builds that answer one word at a time.
Jev is a picker. It works like a multiple-choice quiz. You hand it the question and the answers it's allowed to give. It ticks one box and tells you how sure it is. Same email, and Jev simply says: "Yes, 91% sure."
The company calls this a "System One" model. The name comes from psychologist Daniel Kahneman's famous book Thinking, Fast and Slow. "System 1" is the quick, gut-feeling part of your brain. It's what lets you recognise a friend's face instantly, without having to think it through.
How it works, in three steps
Imagine the post office sorting a big pile of letters. Here's how Jev would help:
- You show it the letter. This is the information it needs to look at: an email, a customer complaint, a form.
- You give it the questions and the possible answers. Which department should get this: billing, technical support or sales? Is the customer angry: yes or no?
- Jev answers all the questions at once, each with a confidence score. The computer program then does the rest: it sends the letter to the right desk, or hands it to a real person if Jev isn't sure.
That last bit matters. Because Jev always says how confident it is, a program can be set up to ask a human whenever the answer is a coin toss.
Why it's so fast and cheap
Writing is slow. When ChatGPT writes a paragraph, it produces one word, then the next, then the next, and the company pays for every one of them.
Jev skips the writing. It reads, thinks, and ticks its boxes in a single step. According to TypeSafe, Jev answers in 0.07 to 0.5 seconds, compared with several seconds or more for typical chat AIs. In one side-by-side test on the company's website, Jev answered in about a tenth of a second, while a well-known chat model took more than eight.
The price is tiny too. Reading costs $42 per billion "tokens" (a token is roughly a piece of a word), and the answers are free.
To show off, TypeSafe let Jev play the old 1990s computer game Doom. It made 10 decisions every second, at a running cost of roughly $7 an hour.
Will I ever use Jev?
Probably not directly, and that's the point. You won't chat with Jev on your tablet. It's a tool for the people who build apps and websites, and it works quietly behind the scenes.
But you may well benefit from it. Think of the things computers already sort for you: spam filters, the bank flagging an odd payment, a customer-service inbox deciding who gets your complaint. Jev is built to make those small, everyday decisions faster, cheaper and more consistent.
Why this matters
- It can't ramble. Jev can only answer with one of the options it was given. It can't invent an answer out of thin air, the way chatbots sometimes do.
- It knows when it's unsure. Every answer comes with a confidence score, so a human can step in on the tricky cases.
- Cheap decisions mean more of them. The name Jev comes from the 19th-century economist William Stanley Jevons. He noticed that when steam engines used coal more efficiently, people ended up using more coal, not less. TypeSafe expects the same with cheap AI.
What it can't do
Jev can't write you a letter, hold a conversation, or do arithmetic. TypeSafe itself advises letting ordinary computer code handle the maths. And picking from a list doesn't mean always picking right: Jev can still choose the wrong box.
The headline claims, like being up to 193 times faster and 444 times cheaper than big chat models, come from TypeSafe's own tests. The company says those are the best-case results, not the typical ones. Independent experts are only just starting to check them.
The takeaway: Jev is the quiet clerk to ChatGPT's chatty writer. It won't keep you company, but it could make a lot of the technology around you snappier and more reliable, one ticked box at a time.
Sources: TypeSafe AI announcement · VKTR · The Register
