OpenAI has introduced Ultrafast, a new processing mode for its flagship model GPT-5.6 Sol, claiming...

OpenAI has introduced Ultrafast, a new processing mode for its flagship model GPT-5.6 Sol, claiming it operates at 14 times the speed of standard processing. The mode generates up to 750 output tokens per second, enabling near-real-time responses. Ultrafast is currently in preview and powered by a partnership with chipmaker Cerebras, with access initially limited to a small group of customers. OpenAI plans to expand availability as capacity increases.

This announcement signals a shift in the competitive landscape of AI inference speed. While Anthropic has launched a fast mode for Claude, it does not match the throughput OpenAI is offering. Ultrafast positions OpenAI to serve latency-sensitive enterprise workflows such as incident response, customer service, financial market analysis, and e-commerce, where real-time performance is critical. The Cerebras partnership underscores the growing importance of specialized silicon in delivering speed advantages that general-purpose GPUs may not provide.

For builders and investors, Ultrafast suggests that inference efficiency is becoming a key differentiator for frontier model adoption. Enterprises evaluating AI vendors should consider throughput and latency as decisive factors beyond raw model capability. The use of Cerebras hardware also highlights a trend toward alternative compute architectures for inference, potentially reshaping infrastructure decisions. As OpenAI expands access, expect competitors to respond with speed-focused offerings or partnerships to close the gap.