PaperBench: Evaluating AI’s Ability to Replicate AI ResearchBy OpenAI News / April 2, 2025 We introduce PaperBench, a benchmark evaluating the ability of AI agents to replicate state-of-the-art AI research.