About TierCliff
An independent calculator for OpenAI service tiers, the 272K long-context cliff and what each lever does to a bill.
TierCliff is an independent, single-purpose tool: it tells you which OpenAI service tier a request should use, and what that choice costs. It was built and is maintained by Charles, a developer who kept finding the same thing in API bills — the biggest cost levers are the ones nobody sets.
The site is not affiliated with, endorsed by, or sponsored by OpenAI. It has no accounts, no signup and no server-side storage of anything you type.
Why this site exists
OpenAI prices a request along three axes at once, and the interactions between them are where money is lost:
- The service tier. Batch and Flex are both half price. Fast is double. Ultrafast is six times. These are multipliers on the same token rates, and the cheapest one is the one teams forget to set.
- The long-context threshold. Cross 272,000 input tokens and the entire request is re-priced — input, cached input, cache writes and output all move. It is a cliff, not a slope, and it is easy to cross by accident when an agent loop appends history every turn.
- Prompt caching. Cached input reads far below the standard input rate — a tenth of it on most models, and a twentieth on GPT-6.1 Sol. Under long context both sides double, so the relative saving survives, but the cache-write cost changes the break-even point.
Each of those is documented somewhere. What is missing is a place that multiplies them together for a specific request, which is what the calculator does. It also prints the multipliers it applied, so you can check the arithmetic rather than trust it.
What makes it different from a price table
A price table gives you a rate per million tokens at Standard tier, short context. It does not tell you whether your overnight enrichment job should be on Batch, whether your retrieval step is quietly over 272K, or whether buying Fast for one interactive endpoint is worth more than the latency it saves.
Every number on this site is traceable to one of two places:
- Published rates and multipliers, read from OpenAI's own API pricing documentation and dated in the footer.
- Planning assumptions, which are labelled as assumptions wherever they appear and which you can override in the calculator.
You can read the full breakdown, including the three figures we have not been able to confirm, on the methodology page.
How it is funded
The site carries advertising. That is the whole business model — there is no paid tier, no affiliate arrangement with OpenAI, and no sponsored content. Advertising has no influence on which tier the tool recommends for a given request; those recommendations follow from the arithmetic, and you can check the arithmetic yourself.
Editorial rules
- Prices are re-verified against OpenAI's official pricing documentation, and the verification date is published in the footer of every page rather than buried in a changelog.
- Where a figure is an assumption rather than a published rate, it is labelled as an assumption in the same paragraph it appears in.
- Where a figure is unconfirmed, it says so. The methodology page lists the open questions rather than papering over them.
- Nothing is published that cannot be recomputed from the numbers shown. If a table on this site disagrees with the calculator, that is a bug, and we want to hear about it.
Getting in touch
Corrections, disagreements about methodology and questions all go to the same place. See the contact page.
Where to start
- The calculator, for a per-request and monthly number on your own workload
- Service tiers explained, for what each tier costs and when it applies
- The 272K cliff, for the threshold that re-prices a whole request
- Guides, for the longer write-ups on Batch, Flex, caching and cost control