Jev, an AI model that makes decisions instead of writing text
TypeSafe AI opens early access to a model built for software automation rather than conversation.
Diogo Almeida, a researcher who worked at OpenAI, has launched Jev, the first public model from his startup TypeSafe AI. Unlike a chatbot, Jev does not generate text: it takes decisions directly in place of code.
The model uses an RLCD training method. TypeSafe AI presents it as extremely fast to respond, far cheaper to run and free of the digital hallucinations that affect text-generating models. Almeida contributed to the instruction fine-tuning methods behind ChatGPT, and says he wondered why such models had not produced massive automation.
Early access is open, aimed at software automation. The applications put forward include classification, scoring and routing requests.
The launch video, published on 15 September 2026, passed 10 million views in about ten hours, and the accompanying blog post reached the top of Hacker News.
