Grok 4.5 cuts output efficiency to a quarter of Opus, reshaping agent economics

xAI's Grok 4.5 uses 15,954 output tokens per task versus Opus's 67,020—a 4.2x reduction. For agents that chain multiple tool calls, this token efficiency gap reshapes operating costs at scale, independent of raw benchmark performance. The arithmetic of running these systems changes.