TypeSafe AI has launched Jev, an AI model that gives calibrated decisions instead of text, cutting hallucinations and costs. The model was launched on September 15, 2026, led by Diogo Almeida, a former OpenAI researcher who helped build ChatGPT and was a co-creator of Reinforcement Learning from Human Feedback (RLHF). Almeida said he left OpenAI because current models were not very effective for automation tasks. Jev was developed under TypeSafe AI's first System One Model, described as a new class of frontier AI models built to make fast, structured decisions for software automation. Unlike typical AI chatbots, Jev is not a standard large language model (LLM); instead of generating text, it gives "calibrated decisions" based on set probabilities. The company claims Jev makes fewer mistakes, works faster, and costs less to run. Jev costs $0.042 per million input tokens and free output tokens. When a user types a query, Jev receives the State and the Typed Question, analyses the information, and calculates the answer in a quick, single go. Outputs are shown as Noul, Choice, and Score.