OpenAI API pricing calculator

Cost is (input tokens ÷ 1,000,000 × input price) + (output tokens ÷ 1,000,000 × output price), multiplied by the number of requests. Input and output are priced separately and output is always dearer, usually four to eight times. The calculator above does this for you, including the cached and batch rates.

It depends entirely on the model. On current prices, gpt-5-nano is $0.05 per million input tokens while gpt-5.5-pro is $30. For a sense of scale, a support assistant sending 1,200 input and 600 output tokens over 10,000 conversations costs roughly $75 on gpt-5.1 and about $3 on gpt-5-nano.

That is the unit every OpenAI price is quoted in. One million input tokens is $1.25 on gpt-5, $0.25 on gpt-5-mini and $0.05 on gpt-5-nano. Output on the same models is $10, $2 and $0.40. A million tokens is roughly 750,000 English words.

A token is a chunk of text, usually part of a word. In English it averages about four characters, so 1,000 tokens is around 750 words. Both what you send and what the model returns are counted, which is why a chatty system prompt costs money on every single call.

Reading your prompt can be done in parallel; writing the reply has to happen one token at a time, so it occupies the hardware for longer. Output typically costs four to eight times input, which means the cheapest thing you can do is ask for shorter answers.

When consecutive requests start with the same long prefix, OpenAI can reuse its work on that prefix and bills those tokens at a much lower rate, often a tenth. It applies automatically. To benefit, put the parts that never change at the very start of the prompt and the parts that vary at the end.

Half. You submit a file of requests and get results back within 24 hours, at 50% of the standard rate for both input and output. It suits evaluations, backfills, bulk classification and anything overnight. It is not usable for anything a person is waiting on.

Among current text models, gpt-5-nano at $0.05 input and $0.40 output. Before reaching for a bigger one, try the smallest that passes your evals: the gap between the nano and flagship tiers is often more than twentyfold, which is a larger saving than any prompt optimisation will give you.

No. API usage is paid from the first request, separately from any ChatGPT subscription. New accounts sometimes get a small trial credit that expires, and OpenAI has at times offered free tokens in exchange for sharing data, but there is no standing free allowance.

In the order that pays best: move to a smaller model, cap output length, restructure prompts so the fixed part comes first and gets cached, batch anything that can wait 24 hours, and trim the system prompt. Switching model usually beats every prompt tweak combined.

They are billed differently and do not overlap. ChatGPT Plus is a flat monthly fee for a person using the app. The API is metered per token for software you build, and a subscription grants no API credit. Light API use costs far less than a subscription; heavy use costs far more.

The date above the table is when they were last read off OpenAI's own pricing page, and that page is linked next to it. OpenAI changes prices and retires models without notice, so treat this as an estimate and confirm against the source before you commit to a budget.