Most AI models are built to talk to people. Jev is built to talk to software. Released in September by San Francisco startup TypeSafe AI, it does not write sentences at all. It takes information in and returns a decision with a probability attached, such as how likely a bank payment is to be fraudulent.
The idea has caught on fast. Demand has forced TypeSafe to suspend sign ups, The Telegraph reported, and copycat systems appeared within days.
What Jev actually does
TypeSafe calls Jev a System One model, after the fast, intuitive kind of thinking in psychology. Instead of reasoning step by step in text, it picks from a set of outputs the developer defines in advance, which TypeSafe says means it cannot invent an answer outside those options.
That makes it suited to the high volume decisions software makes all day: sorting emails, flagging risky commands, routing work to the right model or checking whether an AI agent is about to do something it should not.
Because it skips text, Jev also avoids the slowest part of a language model, which is writing an answer word by word. The output is a number or a label that other software can use straight away, without having to read a paragraph first.
The man behind it
TypeSafe was founded by Diogo Almeida, a former OpenAI researcher who worked on the techniques that made ChatGPT usable, along with Erik Gafni and Sasha Sheng. Almeida says language models are lightning in a bottle that is not useful for many jobs, because computers speak a different language from people.
The company raised about $56 million in seed funding at a valuation of about $281 million, and reports suggest it is now in talks at a valuation above about $14 billion. Almeida says investors are battering down the door.

Faster and cheaper, with caveats
TypeSafe claims Jev is 194 times faster and 445 times cheaper than the best OpenAI and Claude systems. Early developers quoted by TechCrunch reported smaller but still large gains, of 5 to 18 times the speed of a small OpenAI model and 10 to 20 times cheaper than Gemini for their tasks.
Those numbers come from the company and its early users, not independent tests, and they apply to narrow decisions rather than open ended work. Jev cannot write a report, answer a customer or reason through a problem. It is a different tool, not a replacement for a chatbot.
Why Australian businesses should care
A lot of business AI spending goes on simple decisions made at scale: classifying invoices, checking transactions, triaging support tickets. Paying a frontier model to write a paragraph for each of those is expensive, and models like Jev point to a cheaper way, especially in finance automation where fraud and approval checks run constantly.
The sensible move is to test before switching. Jev is still in limited access, and its claims need checking on your own data. Our comparison of GPT 6 Astra, Sol and Luna shows how much the right model size already changes the bill.
A crack in the frontier model story
Analysts at HSBC have warned that cheaper systems like Jev could trigger a sell off similar to the DeepSeek shock, which hit Nvidia, AMD and Google shares. That may be overdone, but the question is real: if many business decisions do not need a giant model, the economics of AI change.
For now, Jev is a reminder that the best AI for a task is not always the biggest one.
