OpenAI has introduced Ultrafast mode for its GPT-5.6 Sol model, which the company states can process requests up to 14 times faster than its standard offering. This new mode is currently in a limited preview phase, accessible to a select group of customers via the OpenAI API.
The Ultrafast mode is powered by Cerebras technology, stemming from a partnership focused on achieving ultra-low-latency inference. OpenAI reports that the enhanced speed allows for the generation of up to 750 output tokens per second.
During the preview, participating customers are deploying Ultrafast mode in production environments across various sectors, including coding, e-commerce, financial research, and customer support. OpenAI aims to leverage this program to identify the most impactful applications for such a significant speed increase and to observe how product development evolves when models can respond almost instantaneously.
Internally, OpenAI is utilizing Ultrafast mode for critical functions such as incident response. This includes tasks like analyzing logs and traces, synthesizing conversations, identifying necessary follow-up checks, and assisting in the preparation and validation of fixes. The company notes that faster inference significantly reduces the time required to observe a signal, test a hypothesis, and determine the subsequent action, while human engineers retain responsibility for judgment and deployment decisions.
Additionally, OpenAI is applying Ultrafast mode to its internal research workflows. This involves tasks such as knowledge searches, data queries, integration with connected tools, and information synthesis. The company suggests that research teams, who might typically launch experiments overnight and review results the next day, can now complete multiple iterations within a single workday.
