Industry Odisha Bureau, Sept 22: A new AI model called Jev, is drawing attention for its approach to AI-decision making and its low cost. Startup TypeSafe AI It has developed it. Unlike conventional large language models (LLMs) that generate text, Jev produces probabilities and what its creators call “calibrated decisions.”
Launched on September 15, Jev is designed to help AI agents decide which tool to use next or whether to retry a task. TypeSafe AI says it can answer such questions in under half a second at a cost of $0.042 per million input tokens, while the output is free.
The model has already been integrated into platforms including Vercel, Cloudflare, LangChain and Langfuse. TypeSafe AI said on September 21 that Jev is now available without a waitlist, with access starting at $5 in credits, equivalent to about 120 million tokens.
Jev was developed by TypeSafe AI co-founders Diogo Almeida, Erik Gafni and Sasha Sheng. Almeida previously worked at OpenAI, where his research contributed to instruction-following technology used in ChatGPT. He also worked on reinforcement learning from human feedback (RLHF), a technique used to improve LLMs.
According to TypeSafe AI, Jev was trained entirely on synthetic data using a method called Reinforcement Learning for Calibrated Decisions. The company described it as a “System One” model focused on intuition rather than entended reasoning.
Instead of generating free form text, Jev returns structured answers and probabilities indicating uncertainty. Its outputs can be used by developers as a decision making layer beneath AI agents.
For other modalities, TypeSafe AI plans to develop additional versions of Jev as well.
