According to Anthropic (@AnthropicAI), API data indicates that Claude AI achieves a 50% success rate on tasks requiring 3.5 hours, with even higher reliability on longer-duration tasks on Claude.ai. These results surpass the typical task horizons found in METR benchmarks, as users can continuously iterate toward a successful outcome on tasks where Claude excels, highlighting significant business opportunities for AI solutions in complex, iterative workflows (Source: AnthropicAI, Jan 15, 2026).