OpenAI Ultrafast makes Astra up to 8x faster in Codex and the API

Ultrafast is a speed setting for Astra, and its headline multiplier compares it with OpenAI's own slower Astra settings, not with other companies' models. In the OpenAI Developers post, the company says that in Codex, Ultrafast runs up to 8 times faster than Astra Standard and 4 times faster than Astra Fast.

The video attached to the post says what kind of speed is being counted, at least for the API. In its narration (an automatic speech transcript; the visuals were not analyzed), the presenter says that in the API, Ultrafast "generates tokens" over seven times faster than the standard tier and over four times faster than the fast tier. Tokens are the small chunks of text a model writes out, so these figures describe how quickly output appears. Neither the post nor the narration says how much sooner a complete task finishes. OpenAI's main launch post, covered separately, also describes the Codex gain as faster token generation, but gives a lower API figure of up to 6 times without naming the baseline, so the API multiplier depends on which announcement and which comparison a reader takes.

The narration also claims that developers no longer have to trade speed for intelligence, and the demonstration generates sprites for a racing game while a regular-speed request is still running. The collected material includes no quality measurement, so the speed claim comes with OpenAI's own figures and the trade-off claim rests on its word alone.