OpenAI has introduced a new high-speed setting called Ultrafast, engineered to dramatically boost the operational pace of its advanced model, GPT-5.6 Sol. The company states that this mode operates at up to 14 times the speed of standard processing, generating as many as 750 output tokens per second.
Powered by OpenAI’s collaboration with chip manufacturer Cerebras, the feature aims to bridge the gap between maximum intelligence and real-time response. Previously, achieving top speed required relying on smaller, specialized models, but Ultrafast demonstrates that top-tier AI can deliver substantial output without sacrificing efficiency.
Currently rolling out in a preview phase for a limited group of users, Ultrafast targets enterprise workflows such as customer support, financial analysis, and incident response. Access will be scaled up gradually as infrastructure capacity expands, keeping pace with similar acceleration efforts from rivals like Anthropic.
- Launch of Ultrafast mode for GPT-5.6 Sol
- Processing speed increased by up to 14x
- Capable of delivering 750 tokens per second
- Built through a technological partnership with Cerebras
- Currently in limited preview for select customers
Sources:
