but the real question is whether any of the 3 models with minimal reasoning outperform gpt-4o and gpt-4.1-mini