Model benchmarks are not perfectly correlated with real-world utility; factors like product quality, writing style, deployment integration, and business model matter more than benchmark performance for determining market success; for example, Anthropic's Claude is often preferred by some users over OpenAI's models despite lower benchmark scores because of writing style preferences.