started · updated
OpenAI launches Ultrafast mode for GPT-5. 6 Sol powered by Cerebras
OpenAI has introduced Ultrafast, a new API service tier designed to run its GPT-5. 6 Sol model at significantly higher speeds. Powered by Cerebras’ Wafer-Scale Engine technology, the mode can generate up to 750 output tokens per second, which is up to 14 times faster than standard processing.
This advancement aims to eliminate the traditional trade-off between model intelligence and response speed. Previously, developers requiring real-time performance often had to opt for smaller, less capable models. Ultrafast allows access to the full reasoning capabilities of GPT-5. 6 Sol with ultra-low latency. The service is currently in a limited preview for a select group of API customers, including companies in sectors such as financial research, coding, customer support, and e-commerce.
In performance benchmarks, Cerebras reported that the Ultrafast mode completed the Humanity’s Last Exam—a 2,500-question PhD-level test—in approximately 11 hours and 11 minutes. This was notably faster than comparable performance from Anthropic’s Claude models. OpenAI is also utilizing the tier internally for time-sensitive tasks like incident response and log analysis.
Entities
Andrew Feldman · Anthropic · Cerebras · GPT-5. 6 Sol · OpenAI · Sachin Katti
Claims
What the coverage asserts, and how well corroborated each claim is across sources.
- [● 18 SOURCES] Ultrafast runs GPT-5. 6 Sol at up to 750 output tokens per second and up to 14× faster than Standard processing. www.globenewswire.com · www.latestly.com · www.unite.ai · noticiasdehoy.com.mx · www.gazetebirlik.com · +13 more
- [● 17 SOURCES] The Ultrafast mode can generate up to 750 output tokens per second. www.globenewswire.com · www.latestly.com · www.unite.ai · noticiasdehoy.com.mx · www.ithome.com · +12 more
- [● 4 SOURCES] In a benchmark called Humanity’s Last Exam, the model completed 2,500 questions in 11 hours and 11 minutes. www.unite.ai · www.blocktempo.com · tbreak.com · www.cryptopolitan.com
- [● 5 SOURCES] Cerebras is powering a new OpenAI API service tier called Ultrafast for the GPT-5. 6 Sol model. www.globenewswire.com · www.latestly.com · www.unite.ai · tbreak.com · www.cryptopolitan.com
- [● 15 SOURCES] The service is initially available in a limited preview to a select group of OpenAI API customers. www.globenewswire.com · www.latestly.com · www.unite.ai · www.gazetebirlik.com · www.ithome.com · +10 more
- [● 4 SOURCES] Cerebras uses Wafer-Scale Engine technology to power the Ultrafast mode. www.blocktempo.com · www.blogspan.net · www.globenewswire.com · www.cryptopolitan.com