Not just personalities and preferences, but the purpose for which the AI is being used also affects the results. I primarily use AIs for complex troubleshooting along the lines of: "Here's a megabyte of logs, an IaC template, and a gibberish error code. What's the reason?" Right now, only Gemini Pro 2.5 has any chance of providing a useful output given those inputs, because its long-context attention is better than any other model's.