GPT-5.6 Sol Ultrafast
Up to 750 tok/s, vendor 14×, not a public GA SKU
OpenAI’s 2026-08-13 previewing-ultrafast post: a limited preview tier powered by Cerebras. Official copy: up to ~750 output tok/s, up to ~14× Standard, same intelligence as GPT-5.6 Sol. Waitlist / limited preview. No separate public model id as a GA SKU. QCode lists gpt-5.6-sol (standard). It does not sell Ultrafast.
Preview facts
Peak output
OpenAI / Cerebras write up to ~750 output tokens per second. Vendor numbers.
vs Standard
OpenAI says up to ~14× Standard. Vendor claim.
Compute
Same-day Cerebras blog: Cerebras powers Ultrafast.
Preview day
openai.com/index/previewing-ultrafast and the Cerebras post landed the same day.
What Ultrafast is
Ultrafast is an OpenAI service tier, not a new model. It runs GPT-5.6 Sol on Cerebras. Vendors say intelligence matches standard Sol; the change is output speed. It launches first as a limited preview in the OpenAI API, behind a waitlist.
No pasteable GA id
The preview has no separate public model id as a GA SKU. Do not invent gpt-5.6-sol-ultrafast and treat it as public GA. The speed lines (750 tok/s, 14×) are OpenAI / Cerebras. This page does not rerun them.
Timeline
GPT-5.6 Sol Standard GA. QCode lists gpt-5.6-sol.
OpenAI previewing-ultrafast. Cerebras blog the same day.
Still limited preview / waitlist. QCode has standard gpt-5.6-sol only. No Ultrafast tier.
Confirmed vs treat carefully
Confirmed
Previewed 2026-08-13; Cerebras-powered; up to ~750 output tok/s; OpenAI says up to ~14× Standard; same intelligence as GPT-5.6 Sol (vendor); waitlist / limited preview; no separate public GA SKU id.
Treat carefully
Do not write Ultrafast as a QCode-buyable tier. Do not invent a public GA model id. Do not treat 750 / 14× as a third-party measurement. Price and a GA date are not published as citable GA figures on the preview page we use here.
Speed preview vs callable now
Ultrafast preview
You specifically want that OpenAI limited-preview speed. Join their waitlist. QCode does not have this tier.
Standard Sol
The same model’s standard surface. QCode can call gpt-5.6-sol. That is not Ultrafast.
How the preview works
Read OpenAI’s previewing-ultrafast page and the same-day Cerebras blog, then use the official waitlist. This is not a GA SKU you drop into model=. Standard Sol is unchanged: gpt-5.6-sol.
On QCode
QCode /models lists gpt-5.6-sol. You can call standard Sol. Ultrafast is an OpenAI-only limited preview, not a QCode tier, and this site does not sell it as Ultrafast. Want that speed? Waitlist at OpenAI. Want Sol today? Use gpt-5.6-sol.
Sol Ultrafast FAQ
Is Ultrafast a new model?
No. It is a service tier that runs GPT-5.6 Sol. Vendors say intelligence is the same.
How fast?
OpenAI / Cerebras write up to ~750 output tok/s; OpenAI says up to ~14× Standard. Vendor numbers.
Same smarts as standard Sol?
Vendors write the same intelligence as GPT-5.6 Sol. We do not rerun that.
What is the public model id?
There is no separate public id as a GA SKU. Do not invent one.
Does QCode have Ultrafast?
No. QCode has standard gpt-5.6-sol. Ultrafast is OpenAI preview only.
How do I get the preview?
OpenAI limited preview / waitlist. The previewing-ultrafast page is the source.
Sources
OpenAI previewing-ultrafast (2026-08-13); same-day Cerebras blog. 750 tok/s and 14× are vendor claims.
Call the Sol that is already listed
Ultrafast is not a QCode tier. Standard gpt-5.6-sol is callable.
Related
GPT-5.6 guide
Sol / Terra / Luna on the GA layer.
Sol vs Terra vs Luna
How to pick a tier. None of them is Ultrafast.
GPT-5.6 pricing
Official standard-tier rates. Do not paste them onto the preview.
Not affiliated with OpenAI / Cerebras. QCode offers gpt-5.6-sol, not Ultrafast. Speed figures are vendor claims.