Preview of Ultrafast mode: GPT-5.6 Sol runs up to 14x faster (up to 750 tokens/s, powered by Cerebras). First rolling out to select OpenAI API customers; built for realtime voice, support, coding, design, finance research and security response.