WTF is Jev!! ⚠️⚠️

Over the past few weeks, many have encountered Jev and experienced misconceptions about its nature. I decided to play with it and the more I used it, the more I realized Jev isn't attempting to be another Claude, GPT or Gemini. Jev is TypeSafe's first System One Model, described by TypeSafe as models built to make fast, structured decisions that software can use directly. System One models are different from traditional AI models that excel at slower, deliberate reasoning.

Jev's approach is to make fast, structured decisions given a state and set of typed questions, returning structured decisions and probabilities. For example, given a customer's state of canceling a subscription, requesting a refund, and already having received two refunds in the last six months, Jev can determine if it's a refund request, assess the risk level, and decide whether to automatically approve or prioritize the request.

The key difference between Jev and other AI models is that Jev's outputs are predefined, and its job is to make the decision. My code then decides what happens next. This architecture differs from LLMs, which are asked to behave like decision engines. Jev's potential outputs are defined beforehand, and my code decides the next steps.

Jev is not just faster than Claude, but it also provides structured decisions and probabilities, making it suitable for specific use cases. TypeSafe reports Jev operating in the tens-to-hundreds of milliseconds range for its System One workloads, with a $0.042/M input-token price at launch. This development could change the way we integrate AI into software, focusing on specific parts that need intelligence instead of relying solely on general-purpose text-generation models.