Math reasoning

GPT-5.2 and Claude Opus 4.6 both score 100% on AIME 2025, while Gemini 3.1 Pro leads GPQA Diamond at 94.3% for PhD-level scientific reasoning.