Cerebras and OpenAI unveiled Ultrafast Mode today, a new service tier in the OpenAI API powered by Cerebras hardware. The Ultrafast Mode enables GPT-5.6 Sol to generate up to 750 output tokens per second without compromising quality. Initially available to a select group of customers, the service aims to accelerate time-sensitive and mission-critical AI workloads, according to cerebras.ai.
The Ultrafast Mode addresses the traditional tradeoff between speed and intelligence in AI models. As models grow larger and more complex, computational demands increase, slowing response times. Cerebras’ solution allows GPT-5.6 Sol to deliver frontier intelligence at unprecedented speeds by optimizing data movement and computation. OpenAI and Cerebras plan to expand access to Ultrafast Mode over time, enhancing AI-powered products and workflows that require rapid, high-quality outputs.
This development is significant amid growing demand for faster AI inference in applications such as real-time decision-making, interactive agents, and large-scale content generation. By enabling GPT-5.6 Sol to operate at ultrafast speeds without quality loss, Cerebras and OpenAI set a new benchmark in AI performance. The move follows broader industry trends where hardware acceleration and software optimization combine to overcome latency bottlenecks in large language models.
Ultrafast Mode’s initial rollout to select customers marks a key milestone in AI service delivery. The OpenAI API documentation confirms the feature’s availability today, with plans for gradual expansion. This launch positions Cerebras as a critical partner in powering next-generation AI models at scale, supporting applications where every second counts.