o1-mini is the compact and economical variant of OpenAI’s o1 series, specially tuned to deliver university-level logic deductions in programming and mathematics with significantly lower latency.
STEM Efficiency and Specialization
By pruning data from non-technical factual encyclopedic knowledge and concentrating neural power exclusively on mathematical and algorithmic reasoning, o1-mini achieves answers up to 80% cheaper than the full o1 model with hardly any loss of technical precision.
AIME 2024
70.0%
Mathematical performance well above traditional models such as GPT-4o (9.3%).
Codeforces
86th percentile
It almost equals o1 standard in solving competitive algorithmic exercises.
Cost
80% Less
Fast inference that makes continuous use viable in IDEs and development tools.
Discover the pricing structure and capabilities of each version in our general guide to OpenAI and ChatGPT.
This post is also available in: