OpenAI announces Ultrafast mode for GPT-5.6 Sol with up to 14x acceleration

OpenAI revealed a preview of Ultrafast mode, a new service level that promises to accelerate GPT-5.6 Sol by up to 14 times conventional processing speed. Initially launched on the company’s API, the technology, developed by Cerebras, is capable of generating up to 750 output tokens per second, integrating OpenAI’s most advanced model into systems and operations where time is crucial.
With this version of GPT-5.6, the company seeks to expand the capabilities of its models, optimizing efficiency across all layers of its infrastructure. Such advances make cutting-edge artificial intelligence more accessible and practical. Historically, achieving real-time speed often required choosing smaller or more specialized models, but Ultrafast technology points to a new phase where intelligence is not sacrificed for agility, enabling more useful work per second.
When speed no longer requires compromising intelligence, AI can be integrated into business sectors that demand high sensitivity to time, opening doors to new ways of working. Several promising scenarios have already been identified for Ultrafast mode:
- Incident response and reliability:In critical systems failures, the mode allows you to analyze application logs, code changes and engineering reports to identify causes and assist in preparing quick solutions.
- Financial research and security:Provides the ability to analyze market signals, evaluate transactions and detect suspicious activity in real time while conditions are still constantly changing.
- Customer support and voice service:Helps resolve complex customer issues instantly, without interrupting the conversation, even if the answer requires multiple steps or systems.
- Business:It makes it possible to answer questions about products, check stock, personalize recommendations and resolve issues at checkout while the consumer decides, avoiding cart abandonment.
- Real-time research and experimentation:It transforms previously all-night studies into interactive work sessions, allowing you to test ideas, analyze results and quickly adjust approaches.
During this preview phase, OpenAI collaborates with an initial group of customers to understand where system speed has the greatest impact and how these observations can improve products in the future. Access expansion will be carried out as service capacity is increased.
Initial experiences of companies with new technology
GPT-5.6 Sol in Ultrafast mode is in the testing phase with a select group of companies active in programming, commerce, financial research, support and other interactive applications. The decision to start with enterprise workflows allows you to study performance in real production environments. This preliminary work is essential to identify where significant acceleration adds the most value and how products transform when artificial intelligence can keep pace with the user, guiding future implementations.
John Crepezzi of Jane Street AI Assistants said, “The speed boost provided by Cerebras is impressive. It enables different ways of using models and makes it practical for developers to work more focused and productively alongside them.”
Internal applications of Ultrafast mode at OpenAI
Internally at OpenAI, a team of developers is also evaluating GPT-5.6 Sol in Ultrafast mode to identify which workflows benefit most from the real-time responsiveness of edge intelligence.
Incident response represents one of the areas where the internal team employs Ultrafast. When faced with an alert, engineers need to create an accurate picture of the situation while the system and evidence continue to evolve. Teams use the tool to quickly read logs, analyze traces, synthesize conversations, determine future checks, and help prepare or validate fixes, all in a fraction of the time, driven by the intelligence of the Sun. This bridges the gap between observing a signal, testing a hypothesis, and deciding the next action, keeping engineers in control of assessment and execution.
In the research field, the team uses Ultrafast to search for sources of knowledge, consult data, as well as collect, organize and summarize information in various interconnected tools. A standard procedure in research involved running experiments overnight and analyzing the results in the morning. With Ultrafast mode, this cycle becomes more agile, allowing multiple iterations even during working hours.
Partnership with Cerebras for the development of innovation
Ultrafast mode symbolizes the advancement in collaboration between OpenAI and Cerebras, with the goal of delivering ultra-low latency inference on the platform. Currently, with GPT-5.6 Sol operating in Ultrafast mode, Cerebras supports OpenAI’s smartest model, guaranteeing up to 750 tokens output per second. This allows companies to develop more responsive products, make decisions more quickly and incorporate powerful artificial intelligence directly into their most critical processes.
When access to Ultrafast will be expanded
The Ultrafast version of GPT-5.6 Sol is currently available in restricted preview for a specific group of customers. OpenAI plans to expand access progressively as system capacity increases.






