Developed by TypeSafe AI, Jev is a specialised decision-making AI model. The company was founded by former OpenAI researcher Diogo Almeida, a co-inventor of RLHF and InstructGPT, alongside Erik Gafni and Sasha Sheng.
Jev replaces standard conversational text - the typical output of LLMs such as Claude and ChatGPT - with raw probability scores tailored for software workflows. Because it skips generating bulky text outputs, it operates dramatically faster and at a fraction of the token cost of frontier models.
Two years ago, Almeida left OpenAI to start TypeSafe AI, a startup trying to fix that problem. This week, the company released a new transformer-based model, Jev, that is not a large language model (LLM). It doesn’t output text, but instead produces probabilities, or what the company calls “calibrated decisions.”
The name Jev was chosen as a nod to the Jevon’s Paradox to reflect the idea that making machine intelligence dramatically cheaper and faster will not decrease AI usage. Instead, it will unlock massive demand, leading to millions of smaller, automated decisions where using a large, expensive language model would previously have been impractical.
Interview with CEO Diogo Almeida on the ideas behind and the possibilities of Jev
How to use
Using Jev with Agents in Microsoft Foundry for Model Evaluation
Further reading
- A new kind of AI model from a ChatGPT inventor is thrilling developers - Techcrunch - 18-Sep-2026
- Jev - Wikipedia
- Jev - Docs
- Jev introduces a new shape of LLM—System One, aka Decision Models - Simon Willison