Introducing Jev: The First System One Model for Automated Decision Making

Research Context and Background

Jev is a new type of artificial intelligence model released by TypeSafe AI. It is the first public model in a category called System One models. Unlike traditional AI chatbots that write sentences one word at a time, Jev is built to make fast, structured decisions that software can use immediately. It was announced on September 15, 2026, to solve the problem of slow and expensive automation in business workflows.

Benefits

Jev offers several major advantages over standard large language models. First, it is incredibly fast. It can answer multiple questions at the same time in a single step, returning results in just 70 to 500 milliseconds. This speed allows it to run in real-time loops, such as controlling a video game character or processing live data streams. Second, it is very cheap. The cost to run Jev is about 40 to 400 times lower than using top-tier chat models for similar tasks. Third, it is highly reliable. Because its answers are limited to specific types like yes or no, numbers, or categories, it cannot make up fake information or break the code that uses it. Finally, every answer includes a confidence score. This lets developers set rules, such as only automating a task if the AI is 95% sure, and sending uncertain cases to a human.

Use Cases

Jev is designed for situations where software needs quick, accurate decisions rather than creative writing. It is perfect for high-volume tasks like sorting thousands of customer support tickets into categories or routing them to the right team. It also works well as a safety check, or guardrail, to quickly verify if the output of another AI model is correct before it is sent to a user. Real-time applications are another strong use case. For example, a video game could use Jev to make a bot react instantly to player moves. It can also handle security incident response by quickly analyzing logs and deciding if an alert needs immediate attention. However, it is not meant for writing emails, generating code, or explaining complex ideas in natural language, as those tasks require the open-ended thinking of traditional chat models.

Pricing

Jev uses a low-cost pricing model designed for automation. The input cost is $0.042 per million tokens, which is roughly $42 per billion tokens. The output cost is free and not metered. This structure makes it viable for processing massive amounts of data or running thousands of decisions per second without breaking the budget.

Vibes

Public reception for Jev has been positive among developers looking for efficient automation tools. Benchmarks show that Jev achieves about 67.8% accuracy on production tasks, which is comparable to mid-tier chat models. While it is slightly less accurate than the very best chat models, it outperforms them in speed and cost. The community has noted that the zero error rate for structured outputs is a major win for building reliable software. Some independent projects have even launched to make Jev accessible through public platforms, showing strong interest in its potential.

Additional Information

Jev is currently available in early access through TypeSafe AI. Developers can sign up for a waitlist or use the API directly to start building. There is also an independent project called OpenJEV that provides public access to the model through OpenRouter. This project funds its operations through trading fees for a token called $JEV. TypeSafe AI defines this new class of models as System One models, drawing a parallel to the human brain's fast, intuitive thinking style to distinguish them from the slow, deliberate reasoning of traditional AI.