Anthropic's Claude Opus 4.6 AI Agents Outperform Human Researchers in Alignment Problem Solving

Nine parallel instances of Anthropic’s Claude Opus 4.6 solved a complex alignment problem with 97% efficiency in five days at a cost of roughly twenty-two dollars per agent-hour, outperforming human research teams on both speed and cost. This result, published by Anthropic on April 15, 2026, marks t

📖 Read the Full Brief