Fastest AI inference API in 2026: 30 days of latency data for Groq, Cerebras, SambaNova and 43 more, from 3 regions