OpenAI’s o3: AI Benchmark Discrepancy Reveals Gaps in Performance Claims Your email has been sent The FrontierMath benchmark from Epoch AI tests generative models on difficult math problems.