xAI's Grok 4.5 scoring 84 on SWE-Atlas-QnA benchmark matches GPT-5.6 Codex for agentic workflows · Digg