OpenAI is previewing Ultrafast, a mode that runs its latest and most capable model, GPT-5.6 Sol, at 14x the speed — per TechCrunch, which frames it as a bid to court enterprise users. The engineering write-up comes from Cerebras, published as "Accelerating GPT-5.6 Sol Ultrafast with OpenAI".
The notable part is the byline, not the multiplier. A frontier OpenAI model served on Cerebras hardware puts wafer-scale silicon in a production path that has been Nvidia's by default, and it makes speed a shipped product tier rather than a benchmark footnote — the same move Groq and Cerebras have been making with open-weight models, now applied to a closed frontier model.
It landed at 220 points and 75 comments on Hacker News inside an hour. Treat the 14x figure as vendor-reported until someone posts independent tokens-per-second numbers; neither source details pricing, availability, or whether output quality is held constant.
Sources: Cerebras, TechCrunch, HN discussion