OpenAI API pricing calculator
Free, instant, no signup. Enter input, output and cached tokens per request plus your monthly volume, and get per-request and monthly GPT cost — with cached input priced at a fifth of the input rate, and each model next to OpenAI's list.
Your workload
Rate: $1.00 in / $6.00 out · cache hit $0.2000 · cache write $1.00 — all per 1M tokens, read live from the Kunavo catalog.
Estimated cost
You save vs OpenAI list
$540/ month · ~60%
Estimate only. Reasoning models bill hidden reasoning tokens at the output rate, so a reasoning workload costs more than the output box suggests. Cache writes are billed separately at $1.00 per 1M and are not in this figure — they are a one-off per cache entry, not per request. Failed requests are never billed.
The two terms that break a naive estimate
Input times rate plus output times rate is a good model of a 2023 bill. On current GPT models it misses in both directions at once.
Cached input pushes the real number down. Any prompt with a stable prefix — a system message, a tool schema, a document you ask five questions about — gets that prefix served from cache at 0.20× the input rate on the OpenAI protocol. A classification job that ships the same 4,000-token instruction with every 800-token record is mostly cache, and a two-term estimate will overstate it badly.
Reasoning tokens push it up, and further. They are billed at the output rate and never returned, so the output field in any calculator — this one included — is the wrong number unless you feed it billed output rather than visible answer length. That is the single most common reason a real invoice beats an estimate.
Where the rates come from
Every rate on this page is read from the Kunavo catalog at render time, the same source the pricing table, the caching docs and the model pages read — and each model is printed alongside OpenAI's own list price so the comparison is checkable rather than claimed. If a rate moves in the catalog it moves here in the same deploy; there are no hand-typed prices on this page.
The Anthropic protocol uses different cache economics — a tenth of input on reads, and no cache-write surcharge on Kunavo — so Claude models have their own calculator rather than a shared one with a filter on it.
OpenAI pricing, in more depth
Every GPT tier, next to OpenAI's list
Subscription tiers vs per-token
How to switch it on, and the rates
Claude, GPT and Gemini side by side
The Anthropic cache rules
Same SDK, different base URL
The endpoint reference
Every model, vs official
How do you calculate OpenAI API cost?
OpenAI bills per million tokens with separate input and output rates, so one request costs (input tokens ÷ 1,000,000 × input rate) + (output tokens ÷ 1,000,000 × output rate), and the monthly figure is that multiplied by request volume. Two terms are missing from that formula on modern models: cached input, billed at a fraction of the input rate, and reasoning tokens, which are billed at the output rate but never appear in the response body.
What is cached input and what does it cost?
Cached input is the leading part of a prompt that repeats across requests — a system prompt, a tool schema, a long document — which the provider can serve from cache instead of re-processing. On Kunavo, cached input for OpenAI-protocol models is billed at 0.20× the model's input rate. It is worth setting the cached field above to a realistic number whenever your prompts share a stable prefix, because the naive two-term estimate will overshoot a repeat-heavy workload substantially.
Why do reasoning models cost more than the output box suggests?
Reasoning models produce internal tokens that are billed at the output rate and are not returned to you. A task whose visible answer is 400 tokens can bill several thousand. This calculator cannot know that ratio for your prompts, so the honest way to use it is to put the total billed output tokens from your own usage data into the output field rather than the length of the answer you see.
Is the OpenAI API cheaper than a ChatGPT subscription?
It depends entirely on volume, and the crossover is easy to compute rather than guess: divide the subscription price by the per-request cost this calculator gives you, and that is how many requests a month the subscription has to save you to break even. Light and bursty usage favours the API; steady daily heavy usage favours a flat subscription. The Codex pricing guide works the same arithmetic through for coding workloads.
Does this calculator count tokens from pasted text?
No — this calculator prices token counts you supply, it does not tokenize text. Character-count estimates drift badly by language and by tokenizer, so the reliable source is the usage object your API responses already return, which reports prompt tokens, completion tokens and the cached portion separately.
Is this OpenAI pricing calculator free?
Yes. It runs entirely in the browser, needs no signup and no API key, and nothing is uploaded. Rates are read live from the Kunavo model catalog and each model is shown next to OpenAI's own published list price, so the comparison can be checked against OpenAI's pricing page rather than taken on trust.