Anthropic's new Claude Opus 4.5 model achieved 80.9% on SWE-bench and scored higher than human candidates on a performance engineering exam. With the AI generating complex code, experts debate if the ...