OpenAI: o3 Mini High

OpenAI o3-mini-high, released on February 12, 2025, is a specialized language model designed for STEM reasoning, building upon the o3-mini model with an elevated reasoning effort setting. It exhibits moderate speed performance, ranking in the 33rd percentile, and is positioned at premium pricing levels, falling into the 14th percentile for cost-efficiency. A standout feature is its exceptional reliability, achieving a 100% success rate across all benchmarks, indicating consistent and stable operation. The model demonstrates strong performance in core STEM areas, achieving perfect accuracy in General Knowledge and Reasoning, and a commendable 92.0% in Mathematics and 90.0% in Coding. These results, particularly the perfect scores, highlight its optimization for complex analytical tasks. While excelling in these domains, its performance in Ethics (98.0% accuracy) and Email Classification (97.0% accuracy) is solid but not top-tier. A notable weakness appears in Instruction Following, where it achieved 62.0% accuracy, suggesting potential limitations with highly complex, multi-layered directives. Its 94.0% accuracy in Hallucinations is respectable, indicating a good ability to acknowledge uncertainty.