Technology

OpenAI O-3

OpenAI's O-3 is a frontier reasoning model: a reflective GPT-class successor to O-1, designed for complex, multi-step problem-solving in technical domains.

O-3 is the next-generation reasoning model from OpenAI, engineered to devote additional deliberation time for step-by-step logical tasks. This reflective generative pre-trained transformer (GPT) architecture significantly improves performance across challenging benchmarks: it achieved 71.7% accuracy on SWE-bench Verified and scored 96.7% on the AIME 2024 math exam, a substantial leap over its predecessor, O-1. The model is available in variants like O-3 Mini (optimized for cost-efficiency) and O-3 Pro (highest capability), allowing developers to adjust reasoning effort (low, medium, high) to balance speed and precision for API-driven applications.