OpenAI is previewing Ultrafast, a new API service tier that runs GPT-5.6 Sol up to 14 times faster than standard processing. The company says the tier generates up to 750 output tokens per second and is powered by chips from Cerebras, extending a partnership that has focused on low-latency inference.
Ultrafast is available in limited preview to a select group of customers, with wider access planned as capacity grows. OpenAI frames the tier as a way to bring frontier intelligence into time-sensitive workflows where waiting on a slower model is not an option.
Early users are testing it in incident response, where engineers read logs and traces while an outage is still unfolding, and in financial research and security, where market signals shift quickly. Customer support and voice agents can resolve multi-step issues mid-conversation, while commerce teams use the speed to answer product questions and catch checkout hesitancy before carts empty.
Inside OpenAI, developers are running experiment loops that once took overnight as interactive sessions, compressing the gap between testing a hypothesis and acting on it. The company says the initial group spans coding, commerce, financial research, and support, and that findings will guide broader deployment.
The move continues OpenAI’s push to close the gap between real-time speed and frontier intelligence, a trade-off that has long forced builders to choose between a small fast model and a big slow one. Ultrafast is OpenAI’s first public answer to that choice for its most capable model.
