Stanford AI Index Report 2026: Gap Between US and Chinese Models Narrows to 2.7%

The Stanford Institute for Human-Centered AI (HAI) released its 2026 AI Index Report, highlighting a shift from pure model capability to systemic changes in infrastructure and organizational processes. Leading models now approach human-level performance on the SWE-bench software engineering benchmark, while agent task success rates in real-world environments surged from 12% to 66%. The performance gap between top US and Chinese models has narrowed to 2.7%, with both sides frequently trading the lead. Additionally, physical infrastructure, including power ca