Why OpenAI models are more efficient than Gemini

OpenAI models achieve higher coding benchmark scores with fewer tokens, compared to Gemini's 250k per task average. The 'Grug speak' theory from leaked GPT 5.5 reasoning traces is suggested as a reason for this efficiency.