# Jev vs GPT-6 Luna, the model behind the OpenAI Decisions API

> GPT-6 Luna, the model OpenAI's new Decisions API is built on, against Jev on 3,600 multi-step reasoning problems: Jev 83.8%, GPT-6 Luna 64.1% on the open-world task. A preview of Luna-based decision accuracy, not a test of the OpenAI Decisions API itself. Results by proof depth with 95% intervals.
