Typesafe AI introduces Jev: a high-speed, low-cost AI model for structured decision-making

San Francisco-based TypeSafe AI has launched Jev, an AI model tailored for answering yes/no and multiple-choice questions with confidence scores. The model is said to be 10 times faster and 10 times cheaper than traditional large language models for decision-making tasks. Jev outputs structured results directly usable by programs, bypassing the need for text generation. The company highlights its efficiency in scenarios like customer service ticket classification, AI agent tool selection, and real-time social media filtering. Jev can tag messages in 2.3 seconds at a cost of 0.1 cents, compared to 20 seconds for chat-based models. It is currently in early trials via API and is positioned as a 'System One' model for structured decisions in software, not for human conversation. TypeSafe AI claims developers can break complex problems into smaller decisions, which Jev can handle efficiently.