The IT world is excited about Jev, a new AI model from Type Safe AI designed to make decisions instead of generating text. Unlike traditional LLMs that output arbitrary strings for tasks like classification, Jev returns structured values from predefined sets—such as 'billing', 'technical', or 'sales'—ensuring reliable, software-consumable outputs. Founded by Diogo Almeida, who previously worked at OpenAI and co-invented RLHF and InstructGPT (key to ChatGPT and GPT-4), Type Safe AI positions Jev as its first System One Model.

Inspired by Daniel Kahneman’s System 1/System 2 framework, Jev delivers fast, intuitive judgments without lengthy reasoning, ideal for real-time software integration. Jev’s cost is striking: input is a fraction of a cent per million tokens, and output tokens are free. This contrasts with using LLMs as decision engines, where inherent randomness risks malformed or unexpected outputs despite structured-output improvements.

Jev eliminates this by constraining the output space to known alternatives, turning classification into a guaranteed, deterministic choice. The model represents a shift from generative AI to decision-focused AI, addressing software needs that require precision over prose. As AI evolves, Jev exemplifies how specialized architectures can outperform general-purpose models in specific, high-reliability tasks.