“OpenAI's strategy with GPT-4 Turbo and GPT-4o involved making the models smaller than their predecessors to reduce serving costs while maintaining or slightly improving quality.”