TypesafeAI, a startup founded by former OpenAI researcher Diogo Almeida, has drawn attention with Jev, an artificial intelligence model designed to make rapid judgments rather than write text, code or converse. Unveiled on September 15, Jev was used by approximately 13% of Vercel’s paid customers within a day, the fastest initial adoption rate among AI models added to the developer cloud platform, Vercel said. Jev classifies unfamiliar inputs according to criteria set by developers, such as shipping, refund, payment or other, and returns a label that backend systems can use to trigger tracking, routing or human support. TypesafeAI says the model can also decide whether an AI agent should answer, forward, continue or stop a task. The company compares its approach with System 1, psychologist Daniel Kahneman’s term for fast, intuitive thinking, and calls Jev a System One Model. In a company-run test of 27 judgment tasks, Jev took 0.114 seconds and cost $0.0001, compared with 8.566 seconds and $0.0139 for an OpenAI model. Those figures imply that Jev was approximately 75 times faster and cost 1/170th as much, although the results have not been independently verified. Unlike Google’s 2018 BERT model, which often requires task-specific fine-tuning, Jev is intended to support multiple tasks through natural-language adjustment of judgment criteria. TypesafeAI has disclosed its architecture and training methods, but not the model’s size, weights or structure. Experts say independent testing is needed, as the wider industry develops AI routing that assigns simple tasks to smaller models and complex work to larger ones.