TypeSafe AI hits $7.5 billion valuation for non-text model Jev

Unlike traditional large language models, Jev operates on a transformer architecture specifically tuned to output calibrated probabilities rather than conversational text. By focusing on computational efficiency, TypeSafe claims the model consumes fewer tokens and executes tasks with greater speed than existing LLMs. The company reports that one-third of Fortune 500 companies have already integrated the system into their workflows.

TypeSafe was established in 2024 by co-founders Diogo Almeida, a former OpenAI researcher, alongside Sasha Sheng and Erik Gafni. Almeida argues that while the industry spent four years perfecting human language processing, that capability remains ill-suited for machine automation. By speaking the language of computers rather than humans, the team aims to capture a distinct segment of the enterprise market that requires high-speed decision logic over creative generation.