OpenAI has launched the Ultrafast mode: GPT-5.6 Sol’s speed has been increased by 14 times and it has entered the preview stage

In August 2026, OpenAI officially launched a new service model called “Ultrafast,” aimed at significantly improving the performance of its latest model GPT-5.6 Sol. This model is driven by collaboration with chip manufacturer Cerebras, enabling GPT-5.6 Sol to process tasks 14 times faster than standard models, with an output token rate of 750 per second. Previously, the industry generally believed that real-time high-speed generation required smaller or dedicated models, but OpenAI stated that Ultrafast represents a new technological direction. Currently, this feature is in the preview stage and available only to a small number of customers. In the future, access will be expanded gradually as capacity increases. OpenAI suggests that this model can be deployed in enterprise workflows such as event response, customer service, financial analysis, and e-commerce. Additionally, GPT-5.6 Sol, as a new generation model, offers stronger capabilities in coding, science, and cybersecurity, and is equipped with the most advanced security stack. It has previously supported integrating workflows into document formats…