OpenAI just gave a preview of something called Ultrafast, and the number attached to it is kind of wild. It runs their GPT-5.6 Sol model up to 14 times faster than the standard version, hitting up to 750 tokens a second. That’s one of the biggest jumps in raw response speed we’ve seen from a frontier model provider this year.
To put that in plain terms, most of us are used to watching an AI type its answer out like it’s thinking through it live. Ultrafast basically skips that whole experience. The response just lands, almost all at once.
The trick behind it isn’t some brand new smarter model. OpenAI built this speed tier on hardware made by Cerebras, whose chips are built to move enormous models through in one pass instead of breaking the work into the usual smaller steps. That’s a different approach than just throwing more GPUs at the problem.
Why does speed like this even matter? Think about anything where a few seconds of delay actually costs something. A trader watching a market move. A support agent on a live call with a frustrated customer. A doctor’s assistant pulling up a patient history mid conversation. Right now, for tasks like that, teams often reach for a smaller, dumber model just because it answers faster. Ultrafast is OpenAI’s bet that you shouldn’t have to trade brains for speed anymore.
It’s only in limited preview for a small group of customers for now, so most of us won’t feel it yet. But if this holds up outside the lab, it changes what people build with AI in the first place. Slow used to be the ceiling for a lot of ideas. That ceiling just moved.
——
Follow: @Ali Demi
Book your free AI clarity call, NOW!
https://buff.ly/TpWy277
——
Sources:
https://openai.com/index/previewing-ultrafast/
https://www.cerebras.ai/blog/accelerating-gpt-5-6-sol-ultrafast-with-openai
https://techcrunch.com/2026/08/13/openai-introduces-ultrafast-a-new-mode-that-makes-gpt-5-6-sol-work-at-14x-the-speed/
Repost this. Thanks.

