“OpenAI has found that using an ensemble of models, where multiple models attempt a problem and a final model integrates their outputs, can produce better results than a single model.”