Hallucination error rate across 26 leading models
0%
50%
100%
22%
best-performing model
94%
worst-performing model
Source: Stanford HAI, The 2026 AI Index Report