[TypeSafe AI Launches Structured Decision Model Jev, Lacks Text Generation Capability] According to the TypeSafe AI blog, the company has released its first 'System One Model,' Jev, which outputs type-safe structured numerical data instead of text, designed for classification, routing, and information extraction. The company states that its end-to-end response time ranges from 70 to 500 milliseconds, compared to 3 to 329 seconds for cutting-edge LLMs on equivalent tasks. Input costs are $0.042 per million tokens, with output being free. The training method is referred to as RLCD (Reinforcement Learning for Calibrated Decisions). Limitations include no image processing, a maximum cardinality of 255 for selected fields, and evaluations based on proprietary business automation workflows rather than public benchmarks like MMLU. Founder Diogo Almeida previously participated in InstructGPT research at OpenAI.