OpenAI launched GPT-6.2 Astra and Sol with Ultrafast, a speed setting that offers up to 4.5x faster token generation in Codex, according to OpenAI’s developer blog. Tokens are the small chunks of text a model writes as it works, so faster generation means code and replies appear sooner. Sol Ultrafast is available on the $200 Pro plan and through the API.
Builds and tests take the same time at any speed
The post’s author uses Ultrafast for interface work, where you tweak a layout, look at it, and ask for another change. In that loop, waiting for the model is the main delay, and a faster model keeps you in the flow.
A task also includes builds, tests, and tool runs, which Ultrafast does not speed up. OpenAI measured an Android timer task in three modes. Standard averaged 16.7 seconds, Fast 16.0, and Ultrafast 15.7, with tests taking about 13 seconds in all three. That is roughly a 6% improvement from Standard to Ultrafast.

The gain is biggest when the model writes a lot of text and your project builds and tests quickly. It is small when most of the time goes to compiling, running tests, or generating images.
API pricing is 3x Standard until February 28, 2027
To use Ultrafast through the Responses API, you set service_tier to ultrafast in the request. OpenAI is charging 3x Standard pricing through February 28, 2027. From March 1, the rate becomes 6x Standard.
Codex’s 0.171.0 release also added GPT-6.2 Sol and Astra to the model catalog, with Sol as the default selection. OpenAI has not said how Ultrafast counts against Pro plan usage limits.
If you plan to build a product on Ultrafast, budget for the 6x rate. A feature that costs three times the normal rate today will cost six times it in March.
End of article