OpenAI Previews Ultrafast Mode for GPT-5.6 Sol, Achieving 14x Speedup on Cerebras Chips

OpenAI has introduced 'Ultrafast,' a new API tier for its flagship GPT-5.6 Sol model, which can run up to 14 times faster, reaching approximately 750 output tokens per second. This significant speed enhancement is achieved by leveraging hardware from wafer-scale chipmaker Cerebras, aiming to reduce latency and make AI agents more genuinely usable. A limited preview of Ultrafast was opened to select customers on August 13, with plans for wider access as capacity expands.