On August 13, OpenAI announced the Ultrafast mode for its latest model, GPT 5.6 Sol. The company claims the new processing mode runs 14 times faster than the standard, capable of generating up to 750 output tokens per second, according to a statement published on TechCrunch.
Tokens are the units of text a language model processes and generates with each interaction. A 14x speed gain in output means that in applications requiring real-time responses — such as customer service chatbots, code assistants, and automation systems — wait times drop from seconds to fractions of a second.
OpenAI did not detail the architectural changes that enable Ultrafast, nor whether the mode requires specific infrastructure, such as dedicated GPUs or specialized load balancing. The announcement suggests the acceleration is an optional configuration, not a new model.
For operations running intensive inference via API, Ultrafast changes the cost-per-request calculation and capacity planning. A 14-fold throughput increase could reduce the number of instances needed to handle the same call volume, but may also imply higher per-token consumption if pricing is not adjusted proportionally. The paper recommends engineering teams measure real latency and total cost before migrating entire workloads to the new mode.