Jev: The AI Model That Decides, Not Chats

Introduction

Jev is a new type of artificial intelligence model created by TypeSafe AI. Unlike most AI tools that are designed to write long paragraphs of text or hold conversations, Jev is built for one main job: making fast and structured decisions. It is the company's first "System One Model." This means it is engineered to give software clear, predictable answers instead of generating explanations. The goal is to solve the problem of using slow, chat-based AI for tasks that simply need a quick yes, no, or a specific choice.

Benefits

Jev offers several key advantages over traditional AI models. First, it is incredibly fast. It can provide answers in as little as 70 milliseconds, which is up to 194 times faster than some other large language models. Second, it is very cheap. The cost is about $0.042 per million input tokens, with no extra charge for the output. This makes it much more affordable than using models like Claude Opus or GPT-5 for simple classification tasks. Third, it is reliable in its format. Jev cannot invent new categories or output messy text that breaks a program. It always returns data in a specific structure that developers expect. Finally, it provides confidence scores. When it makes a choice, it also tells the user how sure it is about that choice.

Use Cases

Jev is designed to work as a specialized part of software workflows rather than replacing general AI chatbots. It acts like a smart decision maker inside a program. One common use is model routing. A system can use Jev to look at a user request and decide which AI model should handle it. It can send simple tasks to a fast, cheap model and save complex reasoning for a powerful, expensive model. Another use is security. Before an AI agent runs a dangerous command, Jev can quickly check if the request is safe or suspicious. It also works well for data classification. It can scan large amounts of data to find sensitive information like medical records or personal details and flag them with a probability score. Developers also use it to test other AI agents by checking if they followed the rules correctly.

Pricing

Jev is currently available in early access. The pricing is advertised at $0.042 per million input tokens. There is no separate charge for output tokens. For example, processing one billion input tokens would cost approximately $42. This is a significant drop compared to the $4,000 to $10,000 that might be required for similar volumes using other frontier models.

Vibes

Public reception has been very positive, with many experts calling it a major shift in how AI is used. Forbes noted that everyone is talking about Jev because it does not chat, which solves a long-standing inefficiency in software automation. The model is seen as a "frontier-intelligence function call" that brings real utility to rigid software systems. While some caution that it can still make factual errors or be vulnerable to prompt injection attacks, the consensus is that its ability to provide structured, probabilistic outputs at a fraction of the cost marks a new frontier in AI utility.

Additional Information

Jev was developed by TypeSafe AI, a company founded by Diogo Almeida. Almeida is a former researcher at OpenAI who co-invented RLHF and InstructGPT. The model uses a training method called Reinforcement Learning for Calibrated Decisions, or RLCD. This method ensures that the probability estimates the model gives reflect true uncertainty. The technology is currently in early access, and developers are encouraged to use robust security measures when integrating it into high-stakes applications.