How Anthropic’s Claude Opus 4.6 Broke Its Own AI Benchmark WinBuzzer