GPT-5.2 reaches expert level in peer review

GPT-5.2 achieved expert-level performance in peer review, with 45 scientists evaluating 469 hours of human and AI reviews across 82 papers and finding the model competitive with top-rated Nature reviewers. The result underscores rapid AI capability gains—from failing basic character counting in June 2024 to solving an 80-year-old combinatorial geometry problem by May 2026—though researchers recommend hybrid human-AI review processes.