New AI tool makes decisions instead of chatting

On September 15, 2026, the company TypeSafe released its first AI model for general use. It's called Jev and works completely differently from the chatbots most people know, such as ChatGPT.

Where ChatGPT is built to write text and answer questions, Jev is built to help software make decisions in the background. That could be, for example, sorting data into categories, routing a request to the right place in a system, or assessing how good a result is. Tasks like these typically happen invisibly to ordinary users, but are used constantly inside the programs and apps we use.

The man behind the model

The man behind TypeSafe is Diogo Almeida. He previously did research at OpenAI, the company behind ChatGPT, where he helped develop a method called RLHF, which among other things made ChatGPT better at following instructions. According to several sources, he has spent two years secretly building a model that speaks more directly to computers instead of guessing its way forward with text, the way ordinary AI models do.

Jev is trained with a new method TypeSafe calls RLCD. In short, the model is meant to give honest probabilities of whether an answer is correct, instead of just producing an answer it isn't sure about. Because the answers always follow a fixed, predetermined format, TypeSafe says the model can't make up wrong answers the way chatbots sometimes do, a phenomenon known as hallucinating.

Faster and cheaper

On speed and price, the sources cite slightly different figures, but they point in the same direction. TypeSafe itself says Jev is 20 to 200 times faster and 40 to 400 times cheaper than comparable models. Other reports round it to about 100 times faster and 200 times cheaper. Response time is 70 to 500 milliseconds, meaning under half a second, compared to several seconds for ordinary language models. The price is set at $0.042 per million pieces of text, so-called tokens, that the model reads, while the tokens it produces itself are, according to TypeSafe, so cheap that they are effectively free.

Several companies have already tested Jev. According to the outlet TechCrunch, the tech company Vercel measured 5 to 18 times faster results than a comparable OpenAI model for security checks. The AI company Bryo found Jev 10 to 20 times cheaper than Google's Gemini model for fast probability calculations.

Access and waiting time

Right now, you can only get access to Jev by signing up for a waitlist at typesafe.ai. After that, you're given a so-called API key, a code that grants access to use the model in your own software. According to the outlet InfoWorld, the service also currently runs from only a single location in the world, and TechCrunch writes that interest after the launch was so great that TypeSafe temporarily ran out of computing power.

None of the sources say anything about when or whether Jev will come to Denmark or the rest of the EU. Companies considering using the model should therefore be prepared to wait and check with TypeSafe themselves whether and when European access becomes an option.

What it means for you

Jev isn't something you'll chat with yourself: it's developers and companies building with it in their software who benefit from it. If you use apps and systems, you may over time find that things like search, sorting and approvals happen faster. If you work in IT or product development, you can sign up for the waitlist at typesafe.ai and see for yourself whether Jev fits your tasks.