What is Jev, the AI that cannot talk but spooked OpenAI?
Picture this. A researcher who helped build the methods behind ChatGPT leaves OpenAI, vanishes into stealth for two years, and comes back with a model that cannot hold a conversation.
On 15 September that researcher's startup, TypeSafe AI, released Jev, together with a 40 million dollar seed round led by DCVC. Developers piled in so fast that the company briefly could not serve them all.
A model that does not talk
Jev is built on a transformer, like the chatbots, but it is not a language model. It never writes text. It outputs probabilities, which the team calls calibrated decisions. You define the possible answers up front, and it tells you how confident it is in each one. It is trained with a method they call reinforcement learning for calibrated decisions.
Two things fall out of that. Because the answers are fixed in advance, it cannot make things up the way a chatbot can. And because it generates nothing, output is free, while input is metered by the billion tokens rather than the million. The company claims up to 200 times faster and 400 times cheaper than comparable language models on classification tasks. Those are their own numbers, so keep a pinch of salt handy.
So why does a giant like OpenAI care about a model that cannot even write a sentence?
Because the giants built their empires on models that do everything, and a huge share of real business work is dull: classify, route, rate, choose. If a small specialist does those jobs cheaper and faster, it quietly takes the highest volume part of the market, the boring part that pays the bills, while the big models keep the glamorous demos. That is an interpretation, not a confession, but the response speaks loudly. OpenAI shipped its own decision API within weeks, and there is already an open source rival called Laya.
The wider lesson is that small teams are not winning by building a bigger brain. They win by finding the one thing the big brain is wasteful at, whether that is the format of the answer or the price of the tokens, and attacking exactly that. Whether the giants can match the price without hurting their own business is the open question.