TypeSafe AI, the company that created the artificial intelligence model Jev, raised US$870 million at a valuation of US$7.5 billion, less than a month after launching it. Jev writes neither text nor code: it delivers decisions together with their probability of being right. The round was led by Andreessen Horowitz, TechCrunch reported on 9 October 2026.
The round in figures
| Item | Detail |
|---|---|
| Company | TypeSafe AI, founded in 2024 |
| Product | Jev, launched on 15 September 2026 |
| Round size | US$870 million |
| Valuation | US$7.5 billion |
| Lead investor | Andreessen Horowitz |
| Other investors | Sequoia and DCVC |
| Founders | Diogo Almeida (ex-OpenAI), Sasha Sheng (ex-Meta) and Erik Gafni |
The figures come from TechCrunch. The company says a third of Fortune 500 companies already use the model, very fast adoption for a product that is weeks old. That is the company's own claim.
What Jev is and how it differs from ChatGPT
Models like ChatGPT are language models: they take in text and return text. Jev uses a related architecture, but its output is different. Instead of writing an answer, it returns a decision of a defined type, for example approve or reject, or a category, along with a number that indicates how sure it is.
| Language model (like ChatGPT) | Jev | |
|---|---|---|
| What it delivers | Text or code | A decision with its probability |
| Designed for | People who read the answer | Programs that act on the result |
| Example | Drafting an email | Deciding whether a payment is suspicious, with 93% confidence |
| What it promises | Quality of the answer | Speed, lower cost and knowing when it is unsure |
The company calls it a "System One model", a nod to fast, intuitive thinking. On its website it sums it up like this: language models produce words for people; Jev produces decisions that are used like code. The practical idea is that a program can set a threshold: if confidence is high, it acts on its own; if it is low, it escalates the case to a person.
"We have been super good at human language for four years, but it's not useful for automation because computers speak a different language," Diogo Almeida, one of the founders, told TechCrunch.
Why investors are excited
Because of cost. Much of what companies want to automate is not writing, but classifying, checking and deciding, millions of times a day. Doing that with a language model is slow and expensive, because it generates text nobody is going to read. TypeSafe claims Jev works much faster and with far fewer resources. If that is confirmed, it makes cheaper precisely the part of AI that is repeated the most.
What has not yet been proven
- The performance figures are the company's. The speed and savings compared with language models come from its own tests.
- The "calibration" has no known outside verification. That the model is right 90% of the time when it says it is 90% sure is its central promise, and it is exactly what a third party needs to measure.
- Adoption is a stated figure. A third of the Fortune 500 using it does not mean they have it in production: it may be a trial.
- A valuation is not a result. US$7.5 billion is what investors bet the company could be worth, not what it sells today.
What this says about the state of AI
That money is starting to look beyond chat. After years in which almost all investment went to models that converse, a round of this size for a model that produces no text indicates the focus is moving towards automation: machines answering other machines.
To understand the basics, see what a prompt is, the instruction used with language models. And on how technology is changing audiences, YouTube now ties with TV as a source of political news.
Frequently asked questions
What is Jev?
An artificial intelligence model from TypeSafe AI that, instead of generating text, delivers decisions with their probability of being right.
How much is TypeSafe AI worth?
US$7.5 billion, after a US$870 million round led by Andreessen Horowitz, according to TechCrunch.
Does Jev replace ChatGPT?
It does not aim at the same thing. ChatGPT converses and writes; Jev is designed for programs to make automatic decisions.
Who created Jev?
Diogo Almeida, a former OpenAI researcher; Sasha Sheng, a former Meta research engineer; and Erik Gafni.
