TypeSafe AI, a startup founded by former OpenAI researcher Diogo Almeida, recently launched the Transformer model Jev. Centered around probabilistic outputs, the model employs 'calibrated decision-making' technology to optimize software automation processes. Jev effectively avoids hallucination issues, reduces computational costs, and adopts a pricing model that is the opposite of mainstream large language models (LLMs). Test results demonstrate Jev's exceptional performance in processing speed and cost control, making it suitable for various scenarios such as command safety classification and monitoring LLM agent behavior. Its training methods, based on synthetic data and 'reinforcement learning through calibrated decision-making,' currently have no direct competitors. Industry experts believe that Jev's exploration may open up new directions for the development of AI models.