GPT-5.4 Pro solves open FrontierMath problem

GPT-5.4 Pro just crossed the rubicon from assistant to researcher, proving that large-scale reasoning can break through the "data wall" of human knowledge. The model established a new lower bound for Ramsey hypergraphs, a Tier 4 problem requiring thousands of sequential logical steps. Verification via the Lean theorem prover demonstrates a successful synthesis of neural intuition and symbolic rigor, mitigating the "hallucination" problem in high-stakes math. This result validates FrontierMath as the gold standard for testing AGI-level reasoning, as earlier models like o3 were unable to score on these open problems. The shift from solving known problems to discovering new mathematical truths suggests AI is becoming a force multiplier for scientific discovery.