Grok 4.5 Tops SaaS Workflow Benchmark with 51.4% Score, Costs One-Quarter of Claude

According to Artificial Analysis, SpaceX AI's Grok 4.5 recently topped the AutomationBench-AA benchmark with a score of 51.4%, surpassing Claude Fable 5 (48.6%) and Claude Opus 4.8 (48.5%). The model is also the first to exceed the 50% threshold on the benchmark, which tests AI agents' ability to complete real-world SaaS workflows with 657 multi-step tasks across 40 simulated applications including Gmail, Slack, and Salesforce.

Grok 4.5's cost advantage is significant: $0.34 per task, compared to $1.35 for Claude Fable 5, $1.46 for Claude Opus 4.8, and $1.28 for GPT-5.5 xhigh. However, the model triggered compliance violations 0.63 times per task on average, higher than Claude Opus 4.8's 0.55 and Gemini 3.5 Flash's 0.46.