In nine days, the storm has surged 50 times, and the valuation of Jev developers has risen to $10 bi

  It's the first time in the history of the world.

In nine days, from $200 million to $10 billion.

According to The Information, Jev's developer TypeSafe AI is talking about a round.More than $1 billionIt's all about financing.

Some investors are willing toMore than $10 billionThe appraisal enters the field!

Just last week, the company announced a $40 million seed round, led by DCVC, valued at $200 million according to PitchBook.

A little more than a week later, the valuation jumped 50 times.

What's even more disgraceful is that there's not a word written about this model where a group of investors are struggling to line up to send money.

The founder of TypeSafe, Diogo Almeida, is a great leader.

He has spent his entire career at OpenAI, working at the core of RLHF and InstructGPT.

In general terms, it is the teaching of GPT to understand people's speech and to learn how to act according to instructions.

This set was later directly inspired by ChatGPT and GPT-4, giving light to the entire big-model era.

In 2024, Almeida left OpenAI to start a business with Erik Gafni and Sasha Sheng.

On September 15, they pushed Jev in front of the stage.

Almeida told Forbes: "Today's big models are trying to appeal to humans, but people shouldn't be the only consumers of intelligence.

What does software need? A direct answer.

If you give it an analysis that's not working, and you give it an option and you give it a confidence, it takes it and runs.

So Jev only did three things:Choose an option, score a point, and decide whether or not to do it.Each answer comes with a probability.

For example, a customer service worker arrives on a single flight and is able to judge three things at once: should it be delivered to the payment group or the account group? should it not be transferred? which grade of emergency is lower to higher?

The code gets the result straight down, without having to retrieve the answer from a sentence.

You can think of it as a classroom with only a selection of subjects. One essay is not written, the answer card is painted fast, and each question is marked with a few hundred impressions.

TypeSafe calls it the "System One model" and responds to quick thinking in the human brain.

Jev is doing this for the software packaging.

The name is also a reference to the "Jev's Paradox" in economics: the more efficient the same thing, the more economical it is, the more people use it instead.

Almeida argues that when judging is cheap to almost non-expensive, there is a doubling of the need for judgement in software.

Tokens are priced at $0.042 per million and are free to export.

After the cancellation of the subscription list on September 20, the new users directly sent a $5 quota, amounting to about 120 million tokens.

So why don't investors sit down? Look at the numbers Vercel gives you.

Jev surpassed all comparison models in the first 12 hours after Vercel's AI Gateway.

By the 24th hour, nearly 13 percent of paid teams had used it, which is the proportion of the GPT-5.6 family.Twice as much.It's Fable 5.1.More than six times。

Vercel himself said: "This is the fastest model release in the history of the gateway".

TypeSafe's own tests are more exaggerated. In the same workflow, Jev is faster than language models.194 timesIt's cheap.445 times。

The developer community is also playing a full game.

Vercel engineer Pranit Sharma uses the Jev inspection command to insure safety, which is 5 to 18 times faster than previous models.

When someone plugged it into a browser plugin, it took an average of 380 milliseconds to decide whether or not a tweet should be folded.

And then there was another guy who took a Jev with the GPT-6 Astra and played the last shot in my world cup, running 8:43 seconds. That part of the Jev cost only 1 cent.

The worst was the other brother.

One night and one morning, he built a fully automated trading robot with Jev, while buying and selling data on and off the chain.$316,680.。

The model is quicker to judge, and it can't handle you.

Together AI has released an open-source, small-scale model tev1-4B, based on the Qwen 3.5-4B modulation, priced at $17, and is also free to export.

According to DataCamp, at least seven open-source Jev alternatives have emerged within a week.

Weight can be reproduced in a matter of days, and TypeSafe can actually protect the fortress, more likely the quality of the calibrated data and the number of developers who have already entered.

What is the story that the investor is buying?

Over the past few years, the big models have been working on one thing: how to talk to people.

But things changed when the Agent got away.

Every step of the way, the agent has to make a decision: which tool is the next step?

These judgments are made thousands of times a day, and each time the big model writes an analysis and gives the answer, slowly and costly.

Jev did a simple but important thing:It's about taking "judgment" out of "generation".It's a very fast and inexpensive model.

There is a bigger trend behind this: AI users are changing from humans to software and agents.

People need explanations and conversations, and machines only need a confident answer.

At a valuation of US$10 billion, the bet is that this category will be established.

As to whether Jev can really eat the market, the 13% of the team is willing to pay back after the end of the free period.

From today, the countdown officially begins.