OpenAI adds Ultrafast tier and Pro 500, drops 5-hour limit

Ultrafast is a speed tier that makes OpenAI's models produce output faster. In its launch post, OpenAI calls it a "premium speed tier" offering up to 8 times faster token generation in Codex, where it reaches 300 tokens per second, and up to 6 times faster in the API. Tokens are the small chunks of text a model writes out, so these figures measure how quickly output appears, not how much sooner a whole task is finished.

The post does not name what Ultrafast is 8 times faster than. A companion post from OpenAI Developers, covered separately, says the Codex comparison is against Astra Standard, with a 4 times gain over Astra Fast. An automatic transcript of that post's video narration puts the API gain at over seven times the standard tier, higher than the 6 times stated here, so the API figure depends on which announcement a reader takes.

This post covers only the speed tier. The video attached to it could not be transcribed, and the material collected here says nothing about price, subscription plans or usage limits, so it cannot confirm or explain the Pro 500 plan or the usage-limit change named in the headline.