Diogo Almeida, the researcher who helped build the methods behind ChatGPT, has launched a new kind of AI called Jev from his startup TypeSafe. Unlike chatbots, Jev does not generate text at all; it makes preset decisions inside software with confidence scores attached.
Jev is built for speed and low cost: it responds in 70–500 milliseconds and charges just $42 per billion input tokens, with output effectively free. TypeSafe says this makes it dozens to hundreds of times faster and cheaper than today’s frontier models on suitable tasks.
The system “can’t hallucinate” by design because it only picks from options defined in advance, such as yes/no, multiple choices, or scores on a scale. That makes it more like a smart database function than a conversational coworker, useful for sorting requests, scoring records, or checking other AIs for jailbreaks.
This marks a shift from chat-first AI to decision-first AI, where precision, reliability, and cost matter more than fluent prose. If Jev delivers on its claims, such models could become a standard, invisible layer inside everyday software.
CHEAPER, FASTER, NO HALLUCINATIONS.
Sanjay Sahay
Have a nice evening.