OpenAI Launches Ultrafast API Tier for GPT-5.6 Sol

OpenAI has unveiled a new high-speed service tier called Ultrafast, designed to run its flagship GPT-5.6 Sol model at dramatically increased speeds. The tier, powered by Cerebras' specialized hardware, is now available in a limited preview to select customers, with broader access planned as capacity expands.

The announcement, made on August 13, 2026, positions Ultrafast as a speed-optimized option within the OpenAI API. According to the companies, the service can generate up to 750 output tokens per second, making it up to 14 times faster than the model's standard processing mode. This performance is achieved without compromising the model's intelligence or requiring a switch to a smaller, less capable version.

The speed is enabled by Cerebras' WaferScale Engine architecture, which uses a wafer-sized chip with 44 gigabytes of on-chip SRAM. This design aims to eliminate the data movement bottlenecks common in GPU-based systems, allowing model weights to stay on-chip and tokens to flow uninterrupted through the model layers.

In benchmark tests conducted by Cerebras, GPT-5.6 Sol on Ultrafast mode demonstrated significant advantages. On the Humanity's Last Exam benchmark, which...