TypeSafe AI's 'Jev' Decision Model Hits 1 Trillion Daily Tokens, Spurring Industry Clones
TypeSafe AI launched 'Jev,' a non-generative decision model that categorizes inputs into predefined outputs and scores options, reaching over one trillion tokens processed daily. Founded by former OpenAI researcher Diogo Almeida, the company's approach has prompted tech giants including OpenAI, Databricks, and Cloudflare to quickly release their own decision-focused tools. Jev highlights an emerging shift away from open-ended generative chat models toward low-cost, high-speed, and deterministic decision engines designed specifically for software automation. Its massive early adoption demonstrates strong enterprise demand for predictable AI components that can be safely embedded directly into automated software pipelines. Instead of generating text auto-regressively, Jev relies on a technique called 'reinforcement learning for calibrated decisions' (RLCD) to output calibrated probabilities, numerical scores, or structured choices. TypeSafe AI previously raised $40 million from VC firm DCVC and is reportedly in talks for a new funding round at a valuation exceeding $10 billion.
## BACKGROUND
Generative large language models (LLMs) produce text by predicting the next token, which offers great flexibility for conversational interfaces but can cause unpredictable or unrepeatable behavior in backend applications. In contrast, non-generative decision models evaluate input state data against fixed, typed options to perform fast classification, scoring, and workflow routing. This makes decision models far more suitable for reliable, multi-layered software automation where predictable logic is required.