OpenAI's documentation suggests a theoretical maximum speed of 280 tokens per second for GPT-6.1 Sol in its fast mode