OpenAI began rolling out Ultrafast mode for GPT-6.1 Sol on Thursday, October 8, in the API, Codex, and ChatGPT Work. It accelerates responses from an existing model, with a higher price for speed and different access rules by product. OpenAI's team announced the release in its developer forum, and the official changelog records its arrival.

What Ultrafast changes

Ultrafast is a processing tier, not a new model. The OpenAI API guide says to choose gpt-6.1-sol and set service_tier to ultrafast. The tier already existed for other models and now includes Sol 6.1.

OpenAI's announcement describes the model as approaching Astra-level intelligence at up to eight times the speed of Sol Standard. That is the vendor's token-generation speed claim, not an independent measurement of complete tasks. A workflow that also invokes tools, waits for network responses, and checks the result could see a smaller end-to-end improvement. The Codex and ChatGPT Work changelog identifies the benefit specifically as faster token generation.

API pricing

The official pricing page lists the following per million tokens for short context:

  • Sol 6.1 Standard: $2 input and $10 output.
  • Sol 6.1 Fast: $4 input and $20 output.
  • Sol 6.1 Ultrafast: $12 input and $60 output.

Ultrafast is six times the Standard price for those categories. For long context, the table lists $24 input and $90 output on Ultrafast. Cached input and cache writes have separate rates, and regional endpoints may add 10%. The choice therefore depends on volume, context length, and the value of less waiting.

The API guide says all API users can request the tier, subject to separate rate limits. It recommends WebSockets for agent workflows that make many calls in succession: a persistent connection reduces network overhead that could otherwise mask some inference gains.

Access in Codex and ChatGPT Work

For these products, the official speed guide lists Pro $500, eligible usage-based or credit-based Enterprise agreements, and credit-based Edu plans. Ultrafast starts disabled for Enterprise workspaces, so owners must enable it. Other self-service plans do not get the tier at launch, even with purchased credits.

The cost of usage changes too. Ultrafast draws down included subscription usage at eight times the Standard rate; purchased credits and Enterprise pay-as-you-go usage are charged at six times the Standard rate. These numbers describe billing or allowance consumption, not response speed. Codex used with an API key follows API token prices instead of those credit multipliers. The documentation also confirms inference residency in the United States and Europe (EEA and Switzerland) for Sol 6.1 Ultrafast.

Where speed has to pay off

OpenAI points to incident response, agents navigating apps, and live experiences as examples. In MnzAI Labs' editorial assessment, the useful test is to compare the same workflow in Standard, Fast, and Ultrafast: time to first response, end-to-end duration, cost per completed task, and result quality. Faster generation matters where waiting slows the work; the price makes it necessary to measure how much waiting actually goes away.