OpenAI previews Cerebras-powered GPT-5.6 Sol Ultrafast
OpenAI opens a limited GPT-5.6 Sol Ultrafast preview to selected API customers, powered by Cerebras and generating up to 750 output tokens per second.
What happened
Event details
Ultrafast is a new service tier for GPT-5.6 Sol. OpenAI publicly reports up to 14 times the speed of Standard processing and 750 output tokens per second. Internal teams have used it for incident response and research loops, but the preview reaches only a small number of customers. Public materials do not substantiate a 20× speed claim and do not say the service runs on Jalapeño, so the two hardware events must remain separate.
Assessment
Why it matters
Ultrafast pushes flagship intelligence toward real-time generation without switching to a smaller model, providing an early test of demand and business value for high-intelligence, low-latency inference.
Availability
Access notes
A limited API preview for a small number of customers. The confirmed public ceiling is 14× Standard and 750 output tokens per second; no date has been given for broader capacity.