OpenAI is previewing Ultrafast mode for GPT-5.6 Sol, powered by Cerebras hardware, with output speeds up to 750 tokens per second — up to 14x faster than Standard mode. It's currently available to a small group of API customers.
OpenAI has tested Ultrafast in scenarios like troubleshooting, research, customer support, financial analysis, and developing agents. For agents that need many sequential model calls, even a faster single response can meaningfully cut the time it takes to finish a task.
That's another step up from the Fast mode OpenAI introduced in late July. Fast mode tops out at 2.5x Standard speed; Ultrafast pushes the ceiling to 14x. OpenAI hasn't shared Ultrafast pricing yet, and it hasn't said when the mode might come to ChatGPT.
https://twitter.com/openai/status/2087947721936359705?s=46