1 total
Reasoning models need different prompts than regular GPT models: the official best practices -- simple and direct beats technique stacking, no step-by-step coaching, seven task types worth delegating, and reasoning-item token savings.