Cerebras powers limited preview of OpenAI Ultrafast service
Cerebras Systems Inc. →Cerebras announced that its inference technology powers Ultrafast mode for GPT-5.6 Sol, a service tier in the OpenAI API. The service was initially available in limited preview to selected OpenAI customers, rather than generally available to every user. The companies described a goal of pairing the model's capabilities with lower response latency for demanding work. OpenAI said the initial customer group would help identify where faster inference creates value and inform later expansion. Cerebras attributed the implementation to its wafer-scale architecture, which keeps model weights on the processor instead of repeatedly moving them to off-chip storage. The announcement establishes a deployed preview and planned learning process, while broader capacity expansion remained prospective.