OpenAI 发布 PaperBench，评估 AI 复现前沿研究能力

We introduce PaperBench, a benchmark evaluating the ability of AI agents to replicate state-of-the-art AI research.