Jev AI: The System One Decision Model

Overview

Jev is a specialized artificial intelligence model created by TypeSafe AI. It was launched in September 2026 to solve a specific problem: making fast, reliable decisions within software applications. Unlike traditional AI chatbots that write long paragraphs of text, Jev is designed to instantly evaluate information and return clear, structured answers like a choice, a score, or a yes or no. It acts as an invisible decision layer that helps applications handle repetitive judgments quickly and cheaply.

The name Jev comes from two ideas. First, it uses "System One Thinking," a concept from psychology that describes fast, automatic, and intuitive decisions. Second, it references the Jevons Paradox, an economic idea suggesting that making something cheaper leads to more use of it. TypeSafe AI believes that by making AI decisions very fast and low cost, developers will use them everywhere in their software infrastructure.

Benefits

Jev offers several key advantages over standard large language models. Its main benefit is speed. It can process requests in just 70 to 500 milliseconds, which is 20 to 200 times faster than traditional models for decision tasks. It also saves money. The cost is very low at $0.042 per million input tokens, and output tokens are free because the responses are small and structured.

Another major benefit is reliability. Jev does not guess or make things up about the format of its answers. Developers define exactly what the answer should look like before the model runs. This means the output is always correct and ready for code to use immediately. The model also provides calibrated confidence scores. If the model says it is 80% sure about an answer, it is actually correct about 80% of the time. This helps developers trust the results without needing to double-check every single decision.

Use Cases

Jev is built to be embedded directly into software systems to handle specific automation tasks. One common use case is customer service routing. When a support ticket arrives, Jev can instantly read the message and decide if it belongs to the billing team, the technical team, or the sales team based on its confidence score.

It is also useful for security alert triage. Security teams often face hundreds of alerts. Jev can rapidly look at each one and prioritize the critical threats while filtering out false positives. Another application is invoice validation. The model can check multiple details on a document at once, such as the total amount, the vendor status, and compliance rules.

Jev can also act as a guardrail for other AI agents. It can review the output of a creative AI to ensure it follows company policies before any action is taken. It can also decide if a user request is simple enough to answer immediately or if it is complex enough to send to a larger, more expensive reasoning model. Finally, it can help maintain consistency in creative pipelines, such as ensuring a character in a story stays the same throughout an AI-generated microdrama.

Pricing

Jev is priced at $0.042 per million input tokens. Output tokens are provided free of charge. This pricing model makes it significantly cheaper than running a full large language model for simple classification tasks. Additionally, adding more questions to a single API call incurs minimal additional cost or latency because all questions are answered in parallel.

Vibes

Jev is currently available in selective early access and stable versions like jev-1.13. Newer experimental versions like jev-1.14 are rolling out periodically. The model is designed to be a complement to large language models rather than a replacement. While LLMs handle open-ended text generation, Jev handles the structured decision layer. This division of labor allows software to be both smart and efficient.

Additional Information

Jev was developed by TypeSafe AI, a startup based in San Francisco that was founded in 2024. The company was led by Diogo Almeida, a former researcher at OpenAI who worked on ChatGPT and GPT-4. Before its public launch, TypeSafe AI raised a $40 million seed round led by DCVC. The model is accessed via a standard POST request to a System One API endpoint. It currently accepts text input, and version 1.14 has introduced experimental support for image input.