Build faster with Ultrafast
OpenAI · 2026-09-29 · official · 45,354 views
What's in the video
Description written by Gemini, which watched and listened to the whole video.
Summary
This official OpenAI promotional video introduces the "Ultrafast" speed tier for GPT-6 Astra across the API, ChatGPT Work, and Codex. A presenter demonstrates its speed through real-time asset generation in a custom racing game and a side-by-side API text-generation benchmark.
What is shown
- [00:00 - 0:07] Title card announcing Ultrafast availability for GPT-6 Astra, followed by the presenter introducing the speed tier.
- [00:08 - 0:23] Side-by-side generation test ("Standard" vs. "Ultrafast") in a game engine tool with the prompt: "Generate a pelican on a bike". Ultrafast finishes generating the 3D asset almost immediately, launching straight into a playable racing scene on a suspension bridge while Standard is still processing.
- [00:24 - 0:31] Side-by-side terminal token generation test comparing "Standard" against "Ultrafast", showing Ultrafast outpacing Standard by reaching ~897 tokens while Standard reaches ~143 tokens.
- [00:32 - 0:44] Closing remarks from the presenter on combining high speed with frontier intelligence, concluding with the OpenAI logo.
Claims & numbers
- Ultrafast is available for GPT-6 Astra in the API, ChatGPT Work, and Codex (stated by the presenter).
- In the API, Ultrafast generates tokens over 7 times faster than the Standard tier (stated by the presenter).
- In the API, Ultrafast generates tokens over 4 times faster than the Fast tier (stated by the presenter).
- Eliminates the traditional tradeoff between generation speed and model intelligence (stated by the presenter).
Notable quotes
- "Ultrafast is now available for GPT-6 Astra in the API, ChatGPT Work, and Codex." [00:01]
- "In the API, Ultrafast generates tokens over seven times faster than Standard tier, and over four times faster than Fast tier." [00:24]
- "In the past, you often had to trade off between speed and intelligence. Now with Ultrafast, you can have both." [00:32]
Assessment
This is an official OpenAI marketing demo showcasing the capabilities and speed differences of the Ultrafast inference tier. The demos illustrate real-time asset creation and raw token throughput, though they are tightly scripted marketing demonstrations rather than an in-depth third-party technical benchmark.
Described by gemini-3.8-flash on 2026-09-30 from the video's audio and frames.