OpenAI added an optional Ultrafast service tier for GPT-6.1 Sol in the Responses API on Oct. 8. Developers can select it with service_tier: "ultrafast"; the company says it shortens the pause between small pieces of generated text, called tokens, as an answer streams. The tier changes how the existing model is served, not the model itself.
The lower latency comes at a higher token price. For short-context requests, OpenAI lists Ultrafast at $12 per million input tokens and $60 per million output tokens, compared with $2 and $10 under Standard. Long-context requests use a separate, higher rate schedule. OpenAI’s API changelog and pricing table list the release and rates.
An OpenAI Developers announcement advertises Ultrafast as up to eight times faster than Standard GPT-6.1 Sol, while the API changelog frames the gain as shorter pauses between output tokens and gives no benchmark conditions, leaving open how much faster full responses or tool-using jobs finish.
The guide recommends a persistent WebSocket for agents that make frequent tool calls, since network overhead can reduce the latency gains. Ultrafast also has separate rate limits from Standard and Fast. The premium may suit requests where users value shorter waits between streamed text, but developers will need to compare costs and latency against their own workloads.
Photo: Derrick Coetzee, Front of server racks at NERSC, CC0 1.0.
