What is Jev, and why is everyone talking about it?

On 15 September 2026, Diogo Almeida, a former OpenAI executive and one of the researchers behind InstructGPT and the RLHF techniques that led to the creation of ChatGPT, posted this viral tweet:

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?

I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev

• 20-200x faster
• 40-400x… pic.twitter.com/JSybNG2BKJ

— Diogo Almeida (@CompleteSkeptic) September 15, 2026

He announced that, after two years of secret testing and stealth development, his company TypeSafe AI is publicly launching the Jev AI model, which has the following features:

• It is 20–200 times FASTER, according to TypeSafe benchmarks, than existing LLM models (ChatGPT, Claude and others)!
• It is 40–400 times CHEAPER than existing LLM models ($0.042 per 1 million input tokens, with no charge for output tokens)!
• Because of its nature, it does not make up answers (hallucinations), as all language models (LLMs) do
• It belongs to a new category that TypeSafe calls “System One”: a category of artificial-intelligence models designed to make fast, structured decisions instead of generating text one word at a time.
• It is based on a new training principle for AI models called RLCD (Reinforcement Learning for Calibrated Decisions)
• It is not a chat model, meaning that it does not write text. It supports three basic answer formats: Choice, Score and Noul (yes/no probability). It can therefore make a choice, give a score on a scale and return a probability. For example, it says “YES with 90% confidence”
• It is not multimodal (yet). In other words, it cannot read images, video, PDFs or anything similar. It reads text only

So what is it useful for?

It is a decision model, meaning an AI model that can make code-ready decisions. Many decisions at once, not token by token like conventional AI models, with incredible speed and almost negligible cost. Unlike language-based LLM models (such as ChatGPT and others), it does not communicate through “human” reasoning logic. LLMs are designed in a particular way that forces them to explain their reasoning by generating text. That approach is useful when a report has to be given to a human, but if it is simply a process inside a program, it is wasted time and money, since every word consumes tokens.

So what does Jev do? It takes information in text form and possible answers, evaluates the information and returns the answer directly, even stating how confident it is in it. And because most software is nothing more than a sequence of decisions, Jev can make a program faster, more efficient and more trustworthy. It does not replace an LLM, but it can take over many of the decisions that an LLM used to make until recently and make them faster and more economical.

What it does in practice

Let us look at some examples from the hundreds that users of x.com have posted:

• Browser navigation at incredible speed. Because navigating the internet is nothing more than a sequence of decisions (click here, scroll, click there, and so on)
• Categorising and prioritising support requests. It can handle hundreds of requests at the same time, categorise them and route them to the right department.
• Email management. It can check hundreds of emails in a few seconds and flag which ones are legitimate, which are spam and which are urgent, etc.
• Reading and categorising hundreds of articles, posts and so on. For example, it can evaluate or categorise content based on specific rules and available evidence, such as flagging posts that need fact-checking.

Why it is so important

Jev can be integrated into existing software and make it faster and more reliable, used to carry out complex workflows almost in real time, and open up possibilities that were not feasible until a few days ago because such a fast and affordable decision solution did not exist.

It is certain that in the future we will see other similar AI decision models like Jev. It is very likely that some of them will come from established major players such as Open AI, Google, Anthropic and others.

Beyond software, Jev’s technology could eventually be used in the autonomous driving systems of cars such as Teslas, in humanoid robots (which are expected to flood the global market in the coming years), and in many other applications and devices such as smartphones, smart speakers, wearables and so on.

What happens now?

Almost no one expected such a development. At a time when all the giants of artificial intelligence have entered a race to build the smartest, biggest and most capable language model (LLM), a new company belonging to a former Open AI researcher is shaking up the global software industry with a solution that bridges the gap and will lead to even faster progress. LLMs will remain necessary because they are the ones that create, write and design, while Jev judges, categorises, routes and scores.

At aiEra, we gained early access to Jev from the first day it was made public, and we have already integrated it into a new customer-service Agents application that we will launch in a few days! Gradually, Jev will be integrated into all our solutions, and we will develop other applications based on the technology of Jev and ChatGPT.

The journey into artificial intelligence has just become even more exciting, even faster…

Yiannis Zafiropoulos

Founder of aiera.gr