OpenAI is making a bold wager that the next AI breakthrough will not be a smarter model, but one fast enough to matter while a crisis, call or transaction is still unfolding. Its new Ultrafast tier promises to bring frontier-model capability into moments where delay has long been the limiting factor.

The company has begun a limited API preview of GPT-5.6 Sol Ultrafast, a Cerebras-powered service tier it says runs up to 14 times faster than standard processing and can produce up to 750 output tokens per second. OpenAI frames the release as a break with the usual trade-off: “Until now, getting real-time speed typically meant choosing a smaller or more specialized model.”¹

The immediate targets are high-pressure business workflows. OpenAI says the mode could help engineers read logs and prepare fixes during an outage, let financial teams assess signals as markets move, and allow support systems to resolve complicated requests without interrupting a customer conversation. It also sees a faster research loop: experiments that once ran overnight could be tested, reviewed and adjusted repeatedly during a workday.

That promise rests on more than raw output speed. The company argues that combining responsiveness with GPT-5.6 Sol’s higher-end reasoning could alter product design itself, making voice assistants, commerce tools and live research systems feel less like delayed chat interfaces and more like active collaborators.

Outside OpenAI’s own framing, the practical restriction is clear: Ultrafast remains available only to a select group of customers while capacity expands. The rollout also lands in an increasingly competitive market for accelerated inference, where rival AI providers have introduced faster modes but not, OpenAI claims, at this stated level of throughput. A report on the launch said the preview is aimed initially at areas including incident response, customer support, market analysis and e-commerce.²

For now, the headline is a controlled test rather than a mass release. But OpenAI’s argument is pointed: if companies no longer have to sacrifice intelligence for speed, the most consequential AI work may move from back-office tasks into live operational decisions.