AI Achieves Breakthroughs in Mathematical Reasoning

OpenAI's Astra solved 10 major open math problems across diverse fields, verified with Lean, at just $2K inference cost—a landmark in AI-driven scientific discovery. Separately, GPT-5.6 Sol and Fable 5 proved an open conjecture on best-of-n with a tighter bound and clean derivation, with minimal human input. A new analysis highlights that these models can now produce correct long-horizon reasoning chains, not just single steps, challenging benchmark sufficiency. These milestones signal that frontier models can now produce publishable research in minutes, challenging assumptions about human-led research and impacting cryptography, coding theory, and formal reasoning.