OpenAI is turning speed into a product feature.
The company has introduced Ultrafast, a preview mode for GPT-5.6 Sol that OpenAI says can work at 14x the speed of standard processing. The headline number is hard to miss: up to 750 output tokens per second, according to the report.
That makes this more than another model setting. For enterprise users, speed changes what AI can be asked to do. A chatbot that answers faster is nice. A model that can process, reason, and respond quickly enough to sit inside operational workflows becomes something else entirely.
Speed Is Becoming A Feature, Not A Bonus
The most revealing line in the reported OpenAI framing is not the 14x claim. It is the idea that, until now, getting real-time speed often meant choosing a smaller or more specialized model. In other words, speed usually came with a trade-off. If you wanted a system to move fast, you often had to accept that it might be less capable, narrower, or less useful for complex work.
Ultrafast is being positioned as a challenge to that trade-off. OpenAI’s pitch, as quoted by TechCrunch, is “more useful work per second.” That phrase matters because it reframes AI performance away from abstract benchmarks and toward throughput. Not just how smart the model is. How much usable work it can produce while the business is still waiting.
The preview is also being aimed squarely at enterprise workflows. TechCrunch says OpenAI points to use cases including incident response, customer service and support, financial market analysis, and e-commerce. These are not casual prompt-and-response moments. They are environments where latency can change the value of the system.
In customer support, a slow model can make automation feel broken even when the answer is right. In incident response, time is part of the problem. In financial analysis, stale output can quickly lose relevance. In e-commerce, the difference between immediate guidance and delayed assistance can be the difference between conversion and abandonment.
This is why speed is becoming part of the enterprise AI sales story. Once companies move beyond experimentation, they stop asking only whether a model can complete a task. They start asking whether it can complete the task within the tempo of the business.
The Enterprise AI Race Is Moving Toward Operational Tempo
OpenAI is not alone in trying to make faster model experiences a selling point. TechCrunch notes that Anthropic has also launched accelerated versions of Claude, including a fast mode, though the report says it does not match the speed OpenAI is claiming here. The comparison is useful because it shows where the category is heading: model makers are no longer competing only on intelligence, coding ability, reasoning depth, or price. They are competing on whether their systems can feel instantaneous enough to become infrastructure.
That is a different kind of race. In consumer AI, speed often feels like polish. In enterprise AI, speed can define the use case. The slower the model, the more likely it remains a tool people consult. The faster it gets, the more plausible it becomes as something embedded directly into service desks, monitoring systems, commerce flows, and internal decision layers.
There is also a distribution signal here. Ultrafast is currently being released in preview to a small group of customers, with OpenAI saying access will expand as capacity grows, according to TechCrunch. That caveat is important. If the mode depends on specialized compute capacity, adoption will not simply be a matter of flipping a switch. The infrastructure has to keep up with the product promise.
The Cerebras partnership sits right at the center of that. TechCrunch reports that Ultrafast is powered by OpenAI’s partnership with the chipmaker, which makes the announcement as much about compute strategy as user experience. The visible feature is faster output. The underlying story is that AI companies are looking for hardware and inference advantages that can be packaged as better products.
This is where the broader platform behavior becomes familiar. AI products are moving from novelty interfaces to daily utility layers, and the winners will be judged by whether people can depend on them at scale. That is also why milestones like Google’s Gemini app reaching 1 billion monthly users matter: mass adoption raises expectations for AI to be available, responsive, and useful in everyday contexts, not impressive only in demos.
For brands and marketers, the immediate takeaway is not to chase “faster AI” as a slogan. It is to understand that AI experience quality is starting to include response time, operational fit, and user patience. If customers are being trained to expect real-time AI assistance, slow branded AI experiences will feel outdated quickly.
Ultrafast still has to prove that the numbers translate into reliable work across real enterprise conditions. But the strategic direction is clear. AI speed is becoming part of the product promise, the infrastructure strategy, and the competitive positioning all at once. In that version of enterprise AI, the model that wins is not just the smartest one. It is the one fast enough to become part of the operation itself.