OpenAI unveils Ultrafast mode for GPT-5.6 Sol, claiming 14x speed boost

OpenAI has introduced a new processing mode called Ultrafast that it says can run its latest flagship model, GPT-5.6 Sol, at up to 14 times the speed of standard inference, delivering as many as 750 output tokens per second. The company announced the preview on Thursday, August 13, 2026, positioning it as a significant leap in real-time AI responsiveness for enterprise applications.

What is Ultrafast and how does it work?

Ultrafast is a new inference mode designed specifically for GPT-5.6 Sol, OpenAI’s most advanced model to date. According to the company, the mode leverages a partnership with chipmaker Cerebras to achieve the dramatic speed increase. In a blog post, OpenAI explained that until now, achieving real-time speed typically required selecting a smaller or more specialized model, but Ultrafast represents a new direction: more useful work per second.

The mode is currently in preview and available only to a small group of customers. OpenAI says it will expand access as capacity grows, suggesting that the underlying infrastructure is still being scaled. The company did not specify which customers are part of the initial rollout or how long the preview period will last.

Why does this matter for enterprise users?

For businesses that rely on AI for time-sensitive tasks, the speed increase could be transformative. OpenAI specifically highlights incident response, customer service and support, financial market analysis, and e-commerce as key use cases. In these environments, faster inference can mean quicker threat detection, more natural customer interactions, and real-time market insights that were previously impossible with slower models.

However, the company has not disclosed pricing for Ultrafast, nor whether it will be an add-on to existing API plans. Enterprises considering the mode will need to weigh the benefits of speed against potential costs and availability constraints.

Competitive landscape and industry context

OpenAI is not alone in pursuing faster inference. Anthropic has also introduced a ‘fast mode’ for its Claude model, though OpenAI claims its speed offering is more substantial. The race to reduce latency reflects a broader industry trend: as AI models become more capable, the demand for real-time interaction grows. Cerebras, known for its wafer-scale chips, has been positioning itself as a key player in high-speed AI inference, and this partnership underscores that strategy.

For now, Ultrafast is a preview, and its long-term impact will depend on how quickly OpenAI can scale capacity and how effectively it can maintain the claimed speed without compromising output quality. The announcement also raises questions about energy consumption and cost efficiency, which are increasingly important in AI deployment decisions.

Conclusion

OpenAI’s Ultrafast mode marks a notable step forward in AI inference speed, with potential to reshape enterprise workflows that depend on rapid responses. While the preview is limited, the technology’s promise is clear. As capacity expands, businesses will likely watch closely to see if Ultrafast lives up to its claims and becomes a standard feature in the AI landscape.

FAQs

Q1: What is Ultrafast mode in GPT-5.6 Sol?
Ultrafast is a new processing mode from OpenAI that enables GPT-5.6 Sol to run at up to 14 times the speed of standard inference, delivering up to 750 output tokens per second. It is currently in preview and powered by a partnership with chipmaker Cerebras.

Q2: Who can access Ultrafast mode?
Ultrafast is currently available only to a small group of customers as part of a preview. OpenAI says it will expand access as capacity grows, but no specific timeline has been provided.

Q3: What are the main use cases for Ultrafast?
OpenAI highlights incident response, customer service and support, financial market analysis, and e-commerce as key applications where the speed boost can deliver significant benefits, enabling real-time decision-making and more natural interactions.