OpenAI has recently introduced a preview of UltraFast mode for GPT-5.6 Sol, targeted specifically at enterprise customers. Under this mode, model inference speed increases by as much as 14 times, reaching a peak throughput of 750 tokens per second. Access to UltraFast mode, however, is not automatic – it requires OpenAI’s approval based on the specific enterprise use case. Companies may submit an application, but whether it is granted rests entirely on OpenAI’s evaluation.
Built for Real-Time, Production-Grade Workloads
OpenAI states that UltraFast mode is powered by Cerebras and is currently available exclusively for the GPT-5.6 Sol model. With UltraFast mode enabled, the model’s generation speed reaches a maximum of 750 tokens per second – fourteen times faster than standard mode.
Naturally, not every workload requires this level of performance. OpenAI has designed UltraFast mode specifically for scenarios demanding real-time or near-production responsiveness: real-time voice applications, customer support, business operations, developer agents, financial research, and security research among them.
Limited Preview Capacity and the Application Process
During the preview phase, the compute resources Cerebras can provide remain limited. As a result, OpenAI will evaluate each enterprise application based on the specific use case the customer describes before granting access to UltraFast mode. OpenAI has indicated that it intends to expand coverage progressively as workload demand grows, and that broader API access to UltraFast mode for a wider range of enterprises may follow in the future.
Enterprise customers interested in gaining early access can submit an application through OpenAI’s official UltraFast request form.
Support Our Threat Intelligence
If you find our technology report and cybersecurity news helpful, consider supporting our work.
Buy Me a Coffee
PayPal
