TypeSafe AI Launches Jev, a Decision Model That Skips Text for 40x-200x Speed Gains in Marketing Automation

Former ChatGPT architect Diogo Almeida's TypeSafe AI unveiled Jev on September 15, a radical new class of AI that sacrifices text generation for blazing speed and minimal cost—becoming Vercel's fastest-adopted model in 24 hours. Unlike traditional large language models, Jev returns typed decisions (Choice, Score, or probability) in a single forward pass rather than generating tokens sequentially. Founded by Almeida alongside Erik Gafni and Sasha Sheng and backed by $40 million from DCVC, the startup built Jev using a training method called Reinforcement Learning for Calibrated Decisions (RLCD), which optimizes for honest confidence scores rather than human preference.