TypeSafe AI's Jev: AI Decisions at Minimal Cost

TypeSafe AI launches Jev, a non-autoregressive model dedicated to calibrated probabilistic decisions. The service opens without a waiting list, features very low costs, and claims speeds on the order of hundreds of milliseconds, with integrations already in place.

Immediate Access, Quick Integrations, and Claimed Gains

One week after its launch, Jev has been integrated by Vercel, Cloudflare, LangChain, and Langfuse. There is no need to sign up for a waiting list to access the service, which offers a plan for $5 corresponding to approximately 120 million tokens. According to TypeSafe AI, the model can be up to 100 times faster and cost up to 100 times less than traditional LLMs for certain tasks.

Typed Decisions in 70 to 500 ms and Pricing at $0.042/M Tokens

Jev returns calibrated decisions in 70 to 500 milliseconds, with an announced pricing of $0.042 per million input tokens. Designed for rapid and structured decisions, the model processes states and typed questions, such as choices, scores, or yes/no. Positioned as TypeSafe AI's System-1 model, it avoids token-by-token generation and is presented as non-autoregressive, intended for outputs consumable by machines. It is a classifier dedicated to this type of decision.

Origins, Funding, and Positioning Against LLMs

Diogo Almeida, a former OpenAI employee, announced Jev on September 15 after leaving the company two years earlier to found TypeSafe AI. The company has raised $40 million and presents a model inspired by the Jevons Paradox. Jev is offered as an alternative to LLMs for software automation and for integrating intelligence into code at a lower cost, with targeted use cases such as ticket classification and routing, prospect or content evaluation, and other automated logics. Nicolas Bustamante believes this approach could illustrate the innovator's dilemma in AI and is part of a phase where the foundation of machine learning, such as classifiers, routers, and context triaging, is becoming commoditized. While these tasks are not new, TypeSafe AI is betting on a model built around typed probabilistic decisions. The project is also presented as a harbinger of probabilistic AI. According to the proposed scenarios, Jev could transform AI agent decision-making by making it faster and less expensive, and could become a building block of the agent ecosystem.