Quick Overview
This short video is a technology highlight presented by Matt Wolfe on recent artificial intelligence developments. Wolfe discusses the launch of Jev, a decision-oriented model created by TypeSafe AI, exploring how it operates differently from standard text-generating language models.
Key Points
- 1.TypeSafe AI released a decision focused System One model called Jev.
- 2.Unlike standard large language models that generate text strings, Jev returns type-safe structured values.
- 3.Jev supports three primitive structured output types consisting of choices, scores, and nouls representing true or false values.
- 4.Input tokens for Jev cost 4.2 cents per million tokens or 42 dollars per billion tokens, while output tokens are free.
- 5.Developer Steven Tey used Jev to build a malicious link scanner for the dub.co link shortener service.
- 6.Developers can prompt coding tools like Codex or Claude Code to generate the integration code needed to run Jev.
Summary
Matt Wolfe highlights an emerging artificial intelligence model called Jev, developed by TypeSafe AI as a System One decision model. Wolfe explains that traditional frontier large language models, such as GPT series models, are built to output sequential strings of text. Even when these models are asked to make decisions, they typically formulate their reasoning and logic by generating text or writing code that must subsequently be executed.
In contrast, Jev bypasses text generation entirely and focuses strictly on returning type-safe structured decisions. The system operates on three primary output primitives: choosing an option from a list, scoring state against a rubric, and evaluating whether a statement is true or false using a primitive called a noul. By eliminating the overhead of text generation, Jev produces direct structured values accompanied by calibrated confidence scores.
This structured approach results in significant cost reductions and high speeds. Jev prices input tokens at 0.042 dollars per million tokens, or 42 dollars per billion tokens, while output tokens are free because they are too cheap to meter. On-screen documentation shows end-to-end response times ranging from 70 milliseconds to 500 milliseconds, representing a forty-times to two-hundred-times speed increase over conventional frontier intelligence queries.
Wolfe demonstrates practical adoption by highlighting developer Steven Tey, who integrated Jev into the link shortener service dub.co to identify malicious URLs before users navigate to them. Wolfe notes that developers can use code generation agents like Codex or Claude Code to automatically generate the necessary integration code to deploy Jev within their own applications.
Introduction to Jev and TypeSafe AI
Matt Wolfe introduces Jev, an artificial intelligence model created by TypeSafe AI. While traditional models like GPT generate sequential text strings to arrive at outcomes, Jev bypasses text output and focuses directly on automated decision making.
Structured Output Primitives and Pricing
Jev delivers type-safe structured data instead of unstructured strings. The model operates across three core primitives: selecting a choice from a list, assigning a numeric score, or evaluating a statement as true or false with a noul. Because it avoids verbose text generation, Jev operates at very low costs, charging 4.2 cents per million input tokens with free output tokens.
Real-World Applications and Integration
The speed and low cost of Jev make it practical for rapid classification tasks, such as Steven Tey's tool detecting malicious destination URLs for dub.co. Developers can also use automated coding agents like Codex and Claude Code to write the software that connects to Jev.
The Bottom Line
The video establishes Jev as a specialized System One AI model from TypeSafe AI that replaces long-form text generation with instant, type-safe decision outputs. It demonstrates how eliminating unnecessary text reduces latency and cuts inference costs to near-negligible rates for classification tasks like URL safety detection. The presentation focuses primarily on initial developer use cases and pricing models, leaving broader enterprise adoption and performance benchmarks across larger datasets unaddressed.
FAQ
What is Jev and how does the TypeSafe AI model work?
Jev is a System One AI model developed by TypeSafe AI that skips text generation to output type-safe structured decisions directly.
How do Jev structured outputs differ from standard large language model outputs?
Standard models output unstructured text strings and code sequentially, whereas Jev returns type-safe values such as choices, scores, and true-or-false nouls.
What are the token pricing rates for the Jev model from TypeSafe AI?
Jev charges 0.042 dollars per million input tokens (or 42 dollars per billion tokens), and output tokens are free.
How did Steven Tey implement Jev for the dub.co link shortener?
Steven Tey trained Jev on a list of over ten thousand malicious domains to build a scanner that flags malicious destination URLs.
Which coding assistants can developers use to write integration code for Jev?
Developers can instruct tools like OpenAI Codex or Anthropic Claude Code to generate the software code required to run Jev.
Worth watching for
Software developers, machine learning engineers, and technical product builders looking for low-latency, cost-effective structured decision models for their applications.
- jev
- typesafe-ai
- artificial-intelligence
- large-language-models
- machine-learning
- dub-co