Built by TypeSafe AI founder Diogo Almeida (a co-inventor of the RLHF work behind ChatGPT) and trained with RLCD (Reinforcement Learning for Calibrated Decisions), Jev cannot generate free-form text — it evaluates a state against a fixed set of typed questions and returns one of three output primitives (a Choice, a Score, or a boolean Noul) with a calibrated confidence value, in roughly 100 milliseconds. TypeSafe AI prices it at $42 per billion input tokens with free output tokens, and publishes benchmarks claiming 20-200x faster and 40-400x cheaper than comparable LLM calls on structured tasks like classification, routing, and moderation — claims that are self-tested and have drawn independent scrutiny. Within 48 hours of launch, at least six independent open-source reimplementations appeared, including the browser-based OpenJev.