We tested the latest GPT-5 models against reasoning benchmarks using Only Connect challenges to measure pattern recognition, lateral thinking, and multi-step inference capabilities beyond traditional knowledge-based tests.