GPT-5.2: Science gains, reasoning gaps

GPT-5.2 leads the FrontierScience eval with gains on hard scientific tasks, but still struggles with the open-ended, iterative reasoning required for real-world research.