Composio benchmark finds Grok 4.5 outperforms GPT-5.5 and GLM-5.2 on multi-page GitHub repository credential audits · Digg