GPT-6.1 Sol Ultrafast puts a price on shorter waits

Waiting for the next line of an answer can interrupt a live editing session. OpenAI added GPT-6.1 Sol Ultrafast on October 8, 2026 to reduce the gap between generated output tokens. It is a service tier for the existing model in the Responses API, available to API users subject to limits. Exact changelog entry.

Set the request’s model to gpt-6.1-sol and service_tier to ultrafast. OpenAI recommends persistent WebSockets for repeated tool calls because connection overhead can erode the gain; HTTP is also supported. The tier has separate limits and supports global processing plus US and EU data residency. Configuration and availability.

What is the shorter wait worth?

For short-context requests, Ultrafast costs USD $12 per million uncached input tokens and $60 per million output tokens, versus Standard’s $2 and $10. Cached inputs, cache writes, long context and regional processing have their own rates. Complete pricing table, model pricing conditions.

An illustrative request with 10,000 uncached input tokens and 2,000 output tokens would cost $0.04 on Standard or $0.24 on Ultrafast, before other charges. We did not run it. That difference may be reasonable while a person waits for an interactive result; it needs a stronger justification for work left running in the background.

Faster token delivery also cannot remove time spent fetching files, running external tools or checking an answer. Compare the complete job, including revisions, rather than assuming a token-speed gain applies to every minute of a workflow.

Next action: if you already use the API, choose one familiar task, inspect your organization’s tier limits and compare elapsed time, accepted output and the full bill within your existing budget. Our Sol release coverage explains the model’s earlier launch; the AI-output evaluation guide helps define what counts as acceptable. API rates do not establish a ChatGPT-plan or Codex-app charge.