OpenAI has unveiled Ultrafast, a new preview mode designed to make its latest model, GPT-5.6 Sol, operate at a much higher pace. The company says the system can deliver up to 750 output tokens per second, reaching roughly 14 times the speed of standard processing.
According to OpenAI, the goal is not only faster responses, but also a new benchmark for how much useful work an AI model can complete in real time. The company notes that this approach moves beyond the traditional trade-off between speed and capability.
Ultrafast is being positioned for practical business use cases such as customer support, incident response, financial analysis, and e-commerce operations. OpenAI says the mode is powered through its collaboration with chipmaker Cerebras, highlighting the growing role of specialized hardware in AI performance.
For now, access is limited to a small group of customers in preview, with broader availability expected as capacity expands. The launch also reflects a wider industry trend, as AI developers race to combine stronger models with faster execution.
As speed becomes a defining feature of AI systems, tools like Ultrafast may help shape a future where intelligent software responds more naturally, efficiently, and at scale.
