OpenAI and Cerebras announced a new service called Ultrafast on Thursday. It is a faster way to run OpenAI's GPT-5.6 Sol model.
The new mode is launching first inside the OpenAI API. It is only available to a select group of customers for now.
Ultrafast can produce up to 750 output tokens per second. Tokens are the small pieces of text a model generates as it responds.
OpenAI says this makes Ultrafast about 14 times faster than the standard version of the model. The company says the speed does not come at the cost of quality.
Compared to other AI models, Ultrafast runs faster too. According to Artificial Analysis, it is 11 times faster than Fable 5 and five times faster than Opus 4.8 in fast mode.
How Ultrafast Was Tested
Cerebras tested Ultrafast against other models using a benchmark called Humanity's Last Exam. The test includes 2,500 questions that usually require a PhD level of knowledge in subjects like chemistry, economics, and literature.
GPT-5.6 Sol running on Ultrafast finished all 2,500 questions in 11 hours and 11 minutes. Fable 5 needed 78 hours and 27 minutes to do the same task.
Cerebras also ran a test called GDP-Val, which measures performance on everyday work tasks. Ultrafast delivered a 5.6 times speedup over the standard version with no drop in quality.
Why Speed Matters For This Tool
OpenAI says GPT-5.6 Sol is built to help with tasks like legal briefs, financial models, and engineering reports. Ultrafast is meant to make these tasks faster to complete.
The company points to a few areas where speed could help the most. These include fixing website outages, handling customer service, and reviewing financial data.
Ultrafast could also help security teams respond faster during cyberattacks. OpenAI says quick detection and response can help limit damage during these events.
Rohan Varma, who works on product at OpenAI, said Ultrafast lets AI keep pace with how people think and work. He said the company is excited to see how it changes daily workflows.
OpenAI researcher Jeffrey Wang said tasks that used to take a couple of minutes now finish almost instantly. He said this makes him more productive day to day.
The speed of Ultrafast comes from Cerebras' hardware. The company uses a Wafer-Scale Engine chip that holds 44 GB of memory directly on the chip.
This design keeps the model's weights on the chip instead of moving them back and forth to separate memory. That movement is normally what slows down standard graphics chips during AI tasks.
Anthropic, one of OpenAI's competitors, also has a fast processing option for its Claude models. OpenAI says that mode does not reach the same speed that Ultrafast offers.
Ultrafast is still in a limited preview. OpenAI says access will expand over time as more computing capacity becomes available.
For now, only a small number of customers can use the tool. People interested in trying it can sign up on OpenAI's website for future updates.
