The smartest ChatGPT can now respond 14 times faster: what’s the point?

On August 13, OpenAI opened a limited preview of Ultrafast, a new service tier for its programming interface. The idea comes down to one line: the same GPT-5.6 Sol, the same intelligence, but served up to 14 times faster than standard processing, or up to 750 text units generated each second. So there’s nothing new on the model front.Everything comes down to the hardware that runs it.

A chip the size of a dinner plate to eliminate the wait

The technical feat comes from Cerebras, an American manufacturer that etches its processors onto a full silicon wafer rather than cutting them into individual chips. The result: a complete model fits on a single component instead of being distributed across racks of graphics cards that spend their time exchanging data. It is these internal round trips that create latency, that handful of seconds between the question and the answer. Eliminating them saves time without affecting the quality of the reasoning.

The demonstration highlighted by OpenAI is compelling. From a single written prompt, the model generates a three-dimensional logistics warehouse simulator, with the company running the exercise in parallel in both modes to make the difference visible. Producing a complete application is nothing like answering a question: the model writes code continuously, and each additional unit of text translates directly into less waiting time. That said, the 14-fold factor remains an in-house measurement, with no externally published verification to date.

Previewing Ultrafast mode: GPT-5.6 Sol at up to 14x the speed.

Launching first in the OpenAI API to a select group of customers with expanded access to more businesses as capacity grows.pic.twitter.com/a5dleofiDJ

— OpenAI (@OpenAI) August 13, 2026

Specifically, who will benefit from it (and when)?

Ultrafast starts on the (API) side—the one developers connect to their own tools—and access remains restricted to a select few clients for as long as it takes capacity to catch up. Among the first testers are financial firms such as Jane Street, as well as players in customer relations and accounting. OpenAI is explicitly targeting workstreams where a late response is no longer of any value: IT incident response, debugging, financial research, fraud detection, real-time customer support, voice applications, and e-commerce. In other words: if you use ChatGPT or Codex, you are not affected.

The company also provides the example of its own teams, which remains the best way to understand the benefits. During a technical incident, its developers have activity logs, traces, and internal exchanges analyzed in real time, then prepare and validate fixes, while the final decision remains human. On the research side, experimental loops that used to run all night now allow for several iterations in a single working day. One tester sums up the shift in their own way: speed “completely changes the calling experience for complex work”.

The awkward question remains: price. OpenAI has announced no pricing, but its current rate card gives a fairly clear indication of where it is heading.The company is already selling its speed: its fast mode, renamed at the end of July and boosted to two and a half times the standard pace, is charged at a premium. Conversely, the plans that accept waiting, whether through deferred processing or variable latency, cost half as much. Every second saved therefore already has a price tag, and Ultrafast runs nearly six times faster than this fast mode, which itself is sold as an add-on.

For Cerebras, this comes at just the right time: the manufacturer, which went public this year, was precisely looking to prove that its giant chips served purposes beyond demonstrations. At the same time, OpenAI gains some breathing room in its dependence on Nvidia. For you as a user, the change will be seen elsewhere and later: in voice assistants that no longer leave pauses, and in customer services that respond at the speed of conversation. The race to build intelligence is not over, but the race for response time has officially begun.

👉🏻 Follow tech news in real time: add 01net to your sources on Google, and subscribe to our WhatsApp channel.

Source: OpenAI