Grok 4.5's Split Verdict: Best Agentic Tool Use on the Board, a Doubled Hallucination Rate, and the New Math of Model Trust - July 11, 2026

Grok 4.5's Split Verdict: Best Agentic Tool Use on the Board, a Doubled Hallucination Rate, and the New Math of Model Trust Grok 4.5's first independent report card is in, and it is split down the middle: the best agentic tool use score on the Artificial Analysis board and a number one claim on the SWE marathon, alongside a hallucination rate that doubled from 25 percent to 54 percent. Chris and Laura unpack how a model can know more yet be wrong more often, why grounded professional work a...