OpenAI is expanding access to its Ultrafast API to a broader user base, signaling rising demand for inference compute.

OpenAI appears to be preparing a wider rollout of its Ultrafast API mode around DevDay on September 29, with new references showing up across the OpenAI Platform and API documentation. TestingCatalog spotted a dedicated speed selector being prepared for the Responses API Playground (currently hidden), where developers could choose between Standard, Fast, and Ultrafast processing.

OpenAI has already officially previewed Ultrafast with GPT-5.6 Sol, describing speeds of up to 750 output tokens per second and up to 14× faster inference than Standard. The feature is powered by Cerebras, and access remains limited to selected customers at present.

OpenAI has since released the GPT-6 family, including GPT-6 Sol and GPT-6 Astra. No confirmation has been made that every GPT-6 model will support Ultrafast at launch.

For developers, the main trade-off will likely be economic. Standard, Fast, and Ultrafast modes could allow organizations to choose latency based on each workload's value, reserving higher-cost inference for applications where response time directly affects revenue or productivity.

The feature fits within OpenAI's broader infrastructure strategy. The company announced a 750 MW Cerebras partnership earlier this year, with capacity being deployed in stages through 2028. TestingCatalog has separately spotted new OpenAI Platform onboarding tiers, including an Accelerate option aimed at production workloads. Together, these changes suggest DevDay could focus heavily on API infrastructure, compute tiers, and tools for developers building production AI systems.