Story
August 14, 2026

OpenAI Wants Frontier AI to Think at the Speed of Business

OpenAI is previewing Cerebras-powered Ultrafast mode for GPT-5.6 Sol, promising 14 times standard speed for live business workflows. The catch: access is limited while the company tests where rapid frontier AI delivers real value.

OpenAI is betting that the next AI advantage will not simply be smarter answers, but answers fast enough to shape decisions while they are still being made. Its new Ultrafast mode makes that case—with access tightly controlled for now.

The company on Thursday previewed a Cerebras-powered tier for GPT-5.6 Sol that it says runs up to 14 times faster than standard processing, generating as many as 750 output tokens a second. OpenAI calls it a “new speed class for frontier intelligence,” aimed at products and workflows where delays carry a real cost.

The pitch is explicitly about avoiding an old trade-off. “Until now, getting real-time speed typically meant choosing a smaller or more specialized model,” OpenAI said, arguing that Ultrafast offers “more useful work per second.” It points to incident response, fraud and market analysis, customer support, commerce and research as early use cases—work in which an AI response must arrive before an outage worsens, a caller disengages or a shopper abandons a purchase.

OpenAI says its own engineers are using the mode to scan logs, traces and conversations during live incidents, while retaining human responsibility for judgment and deployment. Early customers offered a similarly enthusiastic reading: Podium’s Courtland Lykins said the speed “completely changes the call experience for the more complex work.”

The outside account is more measured: Ultrafast is still a preview available only to a small group of customers, with broader release dependent on growing capacity. That makes the launch less a mass-market rollout than a test of whether breathtaking inference speed translates into durable business value.

Sam Altman marked the announcement with a spare “/ultrafast,” while his quoted post confirmed that the API rollout begins with select customers and expands as capacity grows. OpenAI president Greg Brockman called it “wild to see Sol at 14x speed,” underscoring the company’s central claim: the race is now as much about latency as intelligence.