OpenAI Previews 'Ultrafast' Mode That Runs GPT-5.6 Sol 14x Faster on Cerebras Chips

OpenAI has previewed Ultrafast, a new service tier in the OpenAI API that runs its flagship GPT-5.6 Sol model up to 14 times faster than standard processing, delivering as many as 750 output tokens per second on hardware built by wafer-scale chipmaker Cerebras, the companies announced August 13.

Frontier Intelligence at Unprecedented Speed

Ultrafast is not a new model but a new way to serve an existing one: Cerebras' dinner-plate-sized processors, cut from a single silicon wafer, let an entire model sit on one chip rather than being split across racks of GPUs that must constantly shuttle data. OpenAI says the tier runs GPT-5.6 Sol with no quality compromise, and early comparisons cited by Cerebras put it roughly 5 times faster than Claude Opus 4.8 in Fast mode and 11 times faster than Claude Fable 5 based on speeds reported by Artificial Analysis. The tier is available in limited preview to a select group of customers, with access to expand as capacity grows.

Beyond the GPU Standard

The launch is the next step in OpenAI's partnership with Cerebras, which signed a ten-billion-dollar deal earlier this year, and marks a quiet step away from total dependence on Nvidia silicon as OpenAI also develops its own custom chips. OpenAI already monetizes speed in tiers — its Fast mode promises up to 2.5x at roughly double the price — and Ultrafast adds a third, faster tier aimed at latency-sensitive workloads such as real-time agents and high-frequency customer applications. Pricing has not been disclosed, and TechCrunch notes Ultrafast may remain a premium option for mission-critical cases rather than a default setting.