Jev is an artificial-intelligence decision model launched by the startup TypeSafe AI in September 2026. Unlike conversational models such as Claude (AI model family) or ChatGPT, Jev does not chat, write code, or produce free-form prose. Given a piece of application state and a list of typed questions, it returns typed answers with calibrated probability scores attached, typically in a few hundred milliseconds. TypeSafe calls it the first "System One Model," a reference to the fast, intuitive "System 1" thinking described in Daniel Kahneman's book Thinking, Fast and Slow.
How it works
Jev is built for structured decision-making rather than open-ended generation. A caller supplies:
- a block of context or state (for example an invoice, a support ticket, or a security alert), and
- one or more typed questions, each with a defined set of possible answers.
Jev returns the chosen answer for each question along with a calibrated probability, so downstream code can branch on both the decision and the model's confidence. Because the output is typed and constrained rather than natural-language text, results can be consumed programmatically without parsing prose, and the model can respond far faster than a token-by-token chat model. Typical uses include classification, scoring, routing, content moderation, and picking the next step in an automated workflow.
Performance claims
At launch TypeSafe reported that, on its internal workflow evaluations, Jev ran about 193 times faster than Claude (AI model family) Sonnet 5 and cost roughly 444 times less than Claude Opus 5 on decision tasks. These figures come from TypeSafe's own evaluation suite, whose workflows were written by TypeSafe's own team, and the company has said the gains represent the high end of expected real-world use rather than a guaranteed average. Independent benchmarking was still limited in the weeks after release, so the headline speed and cost numbers should be read as vendor claims pending third-party verification.
Background
TypeSafe AI is led by Diogo Almeida, a former OpenAI researcher associated with early work on ChatGPT. The company positions Jev as a complement to, not a replacement for, generative models: large language models remain suited to writing and reasoning in prose ("System 2" work), while Jev targets the high-volume, low-latency classification and routing decisions that sit around them.
Reception
Jev drew rapid attention in the AI developer community after launch, with early adopters reporting on cost and latency improvements for tasks like invoice processing, product matching, and customer-service automation. Coverage noted both the novelty of a typed, non-chat "decision" model and the caveat that its most striking performance numbers were self-reported. The number of projects built on Jev was reported to have grown quickly in its first days of availability. Within days, the Indian startup Convai Innovations released Laya, an Apache-2.0 open-weight model in the same "System One" category, positioning itself as the self-hostable alternative.
See also
- Laya - the open-weight (Apache 2.0) counterpart released days later by India's Convai Innovations
- Large language model
- Generative AI
- AI agent