New Delhi: OpenAI has rolled out a new performance mode designed to make its most capable AI system respond dramatically faster. Named Ultrafast, the feature is built specifically for GPT-5.6 Sol, the company’s top-tier model, and reportedly processes information roughly 14 times quicker than the standard version, churning out as many as 750 tokens (units of generated text) every second.
Historically, developers seeking near-instant AI responses had to trade power for speed, often settling for smaller, less capable models. OpenAI is positioning Ultrafast as a way to sidestep that compromise entirely, framing it as evidence that raw model capability and rapid output no longer have to be mutually exclusive.
This isn’t happening in a vacuum, rival labs are chasing the same goal. Anthropic, for instance, already offers a “fast mode” for its Claude models, though OpenAI’s latest offering reportedly outpaces it in raw throughput.
According to OpenAI, the speed gains open the door to time-sensitive use cases that enterprises care about: responding to security incidents in real time, powering faster customer support interactions, running rapid financial analysis, and supporting high-speed e-commerce operations.
The engine behind this leap is a partnership with Cerebras, a chipmaker known for its specialized AI hardware. For now, Ultrafast remains in a limited preview stage, accessible only to a select set of customers, though OpenAI has indicated it plans to widen availability as infrastructure capacity allows.
Previous
