AI Anthropic's Claude Opus 4.7 posts a jarring benchmark regression that has enterprise AI teams asking uncomfortable questions 4 min 350 views