OpenAI launches Ultrafast GPT-5.6 Sol for businesses

OpenAI has introduced Ultrafast mode for GPT-5.6 Sol in the OpenAI API. The new service tier can run the model at up to 14 times the speed of standard processing.

The limited preview is available to a select group of customers. OpenAI is positioning the release as an early test of where lower-latency AI is most useful in business workflows.

Ultrafast is the first release in what OpenAI described as a new speed class for its frontier models. Powered by Cerebras, the service can generate up to 750 output tokens per second, allowing GPT-5.6 Sol to respond quickly enough for tasks that change in real time.

That marks a shift from a common trade-off in AI deployment, where organisations often chose smaller or narrower models when they needed faster responses. OpenAI said improvements across its systems had made its more advanced model efficient enough to handle time-sensitive work without reducing the level of intelligence on offer.

Business uses

OpenAI has been testing the mode with an initial group of companies working in coding, commerce, financial research, support and other interactive applications. The aim is to observe how a sharp increase in model speed changes products and workflows in live production settings.

Among the examples cited were incident response, where engineers review logs, code changes and internal reports while an outage is still developing. OpenAI also pointed to financial research and security tasks such as analysing market signals, assessing transactions and flagging suspicious activity while conditions are still shifting.

Customer support and voice systems are another focus. OpenAI said the model can be used to resolve complex customer issues in real time, including cases where the answer depends on several steps or multiple systems.

In commerce, the mode could help answer product questions, check stock, tailor recommendations and deal with checkout issues while a customer is still making a purchase decision. Faster response times could also turn research and experimentation from overnight batch work into a more interactive process during the working day.

Internal testing

Inside OpenAI, developers have been using the mode to test where near real-time responses from a frontier model make the biggest difference. Incident response has been one of the main internal use cases.

According to the company, engineers use the system when alerts are triggered to read logs, analyse traces, bring together conversations, identify next checks, and help prepare or validate a fix. OpenAI said the tool shortens the time between spotting a signal, testing a theory and deciding what to do next, though engineers remain responsible for judgment and deployment.

Research teams are also using the mode to search knowledge sources, query data, and gather and organise information across connected tools. OpenAI expects one result to be a shorter experimental cycle, with several rounds of testing possible within a single workday rather than relying on overnight runs.

Cerebras link

The launch extends OpenAI's partnership with Cerebras, which provides the infrastructure behind the new tier. Ultrafast represents the next step in that relationship as OpenAI seeks to bring lower-latency inference to its platform.

The announcement highlights how speed is becoming a product category in its own right in the market for advanced AI services. Rather than focusing only on model size or benchmark performance, suppliers are increasingly trying to show they can deliver useful responses quickly enough for live decision-making, operational support and customer interaction.

For businesses, the practical question is whether faster model output changes how people work rather than simply reducing waiting time. OpenAI is using the preview period to learn where an order-of-magnitude improvement in speed produces the greatest value and how products change when the model can keep pace with the user.

Access remains restricted while OpenAI increases capacity. GPT-5.6 Sol on Ultrafast mode is available in limited preview to a select group of customers.