xAI’s Grok 4.5 placed second on the Mercor APEX-SWE coding benchmark, posting a Pass@1 accuracy score of 51.2%, according to Techub, citing CryptoBriefing. The model trailed Anthropic’s Claude Fable 5, which led the ranking with 65.5%. Grok 4.5 was launched in early July 2026. In a separate benchmark focused on automation capability, AutomationBench-AA, the model ranked first with 51% accuracy. xAI also said Grok 4.5 delivers comparable performance while cutting task costs by 4x to 17x versus competing models. The results add another data point to the competition among frontier AI models, with benchmark performance, automation capability, and operating cost all drawing attention from developers and enterprises evaluating model deployment.