Summary & Key Takeaways ​

• Jev is a new category of AI model, dubbed "Decision Models" or "System One models."
• It accepts text inputs but outputs structured data like floating-point numbers.
• Outputs include confidence scores for yes/no questions, choices, and ratings.
• The model is noted for being extremely fast and significantly cheaper than traditional LLMs.
• Jev charges only for input tokens, with output being free.
• It allows for asking multiple questions about a single "state" object.
• This represents a potential shift in how AI is used for structured decision-making.
• Simon Willison highlights its potential impact on AI application design.

Our Commentary ​

Okay, Jev. This is genuinely fascinating. An LLM that doesn't output text but numbers and confidence scores? That's a game-changer for so many applications where we're currently trying to parse structured data out of unstructured text. The speed and cost are just icing on the cake. I've been waiting for something like this, a model specifically designed for structured decision-making rather than just generation. This feels like a real step forward in making AI more directly actionable and less about prompt engineering for specific output formats. I'm excited to see how this evolves.