OpenAI's Ultrafast charges your included ChatGPT usage at 8x and your credits at 6x
Running ChatGPT's Ultrafast mode costs a subscriber more out of the usage their plan includes than out of the credits they have paid for. OpenAI's pricing page for Codex and ChatGPT sets both multipliers against Standard mode for the same model: Ultrafast draws on included subscription usage at 8x, and on purchased credits and Enterprise pay-as-you-go usage at 6x.
Ultrafast is sold inside Codex and ChatGPT Work, which share one pot of usage. Image: OpenAI Image: OpenAI
Fast mode has the same shape at a smaller scale, at 2.5x included usage against 2x credits. Both Ultrafast models, GPT-6 Astra and GPT-6.1 Sol, carry the identical 8x and 6x pair, read at 21:03 UTC on October 8, so the gap belongs to the speed mode rather than to a model.
Ultrafast is the fastest of OpenAI's three speed settings, sold inside Codex and ChatGPT Work and described by the company as reducing the time between generated output tokens. It is not a separate plan and it is not metered in dollars on a subscription: it runs down the same allowance and the same credit balance as everything else, faster.
The gap is a third on Ultrafast and a quarter on Fast
Eight against six is a difference of 33%, and 2.5 against two is 25%. Both run the same way round, with the included allowance as the dearer of the two buckets, and the allowance is also the one OpenAI spends first on Pro $500, with credits taking over once it has run out, the Speed documentation says. The same page restates the pair in prose: Ultrafast "bills purchased credits and Enterprise pay-as-you-go usage at 6x the Standard rate" and "uses included subscription limits at 8x the Standard rate".
OpenAI's summary of its credit rates gives only one of the two numbers. Fast mode "uses 2x the Standard rate where available, and Ultrafast mode uses 6x" for purchased credits and Enterprise pay-as-you-go usage, it says, and adds: "Included subscription usage has different multipliers."
Enterprise workspaces start with the tier switched off
Ultrafast runs in Codex and ChatGPT Work on the $500 Pro plan that arrived at the end of September and on eligible Enterprise and Edu plans. Enterprise workspaces have it off by default, and an owner has to switch it on for selected users or for the whole workspace, with existing per-user spend controls applying to the Ultrafast usage that follows.
Eligible Edu plans run on credits, and legacy Enterprise plans billed on rate limits rather than on usage are not supported at all. An API key takes the multipliers out of the picture: "With an API key, Codex uses API token pricing instead, and ChatGPT credit multipliers don't apply," OpenAI says, which bills the work at the API's own Ultrafast token rates rather than against a subscription. Otherwise the two products share one pot: "ChatGPT Work and Codex share usage. Both use the same pricing, credits, and usage limits."
Standard credit rates set the base
The multipliers act on OpenAI's published Standard credit rates, quoted per 1M tokens. GPT-6.1 Sol costs 50 credits for input, 2.5 credits for cached input and 250 credits for output. GPT-6 Astra costs 250, 25 and 1,250 on the same three lines, and Codex credit billing carries no separate cache-write charge.
OpenAI gives no reason for the difference. It states both multipliers and says nothing about why an allowance a subscriber has paid for in advance is consumed a third faster than credits bought on top of it.
