How this calculation works
Direct API standard token rates. Long-context rates apply to the entire request above 272,000 total prompt tokens. Cache-write categories represent 30-minute and 12-hour policies. Batch is the documented 50% token-rate variant; Fast and Ultrafast are outside this estimator.
Per request USD = Σ(disjoint billed tokens × selected USD per million / 1,000,000); period = requests/day × days × per request + extra period feesA worked example
A fictional workload of one thousand fresh input tokens and two hundred visible output tokens per request costs 0.004 USD per request at the default listed model rate. One thousand requests each day for thirty days gives 120 USD before any extra fee or tax. The rate table shows how selecting another published model changes that result.
Good to know
Independent estimate from the dated official rate card, not an invoice or live tariff. Verify current model availability, billing categories and provider terms. Do not use subscription allowances as API credit. Taxes, negotiated discounts and unentered charges are excluded.
Which OpenAI bill this estimates
Direct API standard token rates. Long-context rates apply to the entire request above 272,000 total prompt tokens. Cache-write categories represent 30-minute and 12-hour policies. Batch is the documented 50% token-rate variant; Fast and Ultrafast are outside this estimator.
A model selected here is a specific hosted price row. It is not a promise of access, a context-window validator or a statement that every deployment of that model costs the same amount. Use the provider that will actually issue your invoice.
The Anthropic Claude API cost calculator covers a separate account or billing model. Compare its own documented scope rather than carrying a quota percentage or hosted rate across providers.
Enter disjoint token categories
Input categories must not overlap. Fresh input excludes cache reads and separately billed cache writes; the output field should contain visible billed completion tokens. Add reasoning tokens only when the provider reports them separately and they are not already included in the output count.
A million tokens is the pricing denominator, not a minimum purchase. For example, one thousand tokens is one thousandth of a million. The calculator keeps fractional dollars internally and displays small per-request charges without rounding them into an apparent free service.
Choose a model and check the applicable tier
The table links each model row to its official source and displays the checked date. The result identifies the selected billing route and any automatic long-context or scheduled promotional transition supported by this snapshot.
A mixed workload needs separate calculations for differently sized requests or different models. One average prompt length can hide expensive threshold crossings; a request just over a tier boundary may price all of its tokens at the higher tier.
Ways to reduce OpenAI API spending
Use actual usage counters, including billed reasoning output. Cache identical eligible prefixes when appropriate, then measure real cache hits rather than treating all repeated words as cached. Batch can suit asynchronous workloads, but is not a subscription discount.
Compare cost per successful task as well as cost per token. A cheap first attempt followed by many repairs can cost more than a stronger model that finishes correctly. Measure a representative sample before forecasting a large daily volume.
Budget for more than the model response
The additional-fee field is one USD total for the entire entered period. Add independently priced tools, grounding, storage or regional charges there after checking their billing units. It is not a per-request amount and does not receive a token batch discount.
The INR result multiplies total USD by your entered exchange-rate assumption. It is not a live foreign-exchange quote and does not add Indian taxes or payment-provider conversion charges automatically. API accounts, console billing and consumer subscriptions remain separate.
Independent estimate disclaimer
Independent estimate from the dated official rate card, not an invoice or live tariff. Verify current model availability, billing categories and provider terms. Do not use subscription allowances as API credit. Taxes, negotiated discounts and unentered charges are excluded.
Rates checked 9 November 2026. Model lists are a researched selection, not an exhaustive live catalog. A later billing date uses only scheduled changes explicitly recorded in the source snapshot; verify any subsequent unscheduled changes before relying on the forecast. No API key is needed or requested.
A few questions, answered
Does a paid chat plan cover these API calls?
Do not assume so. This calculator estimates the selected API tariff. Any explicit account credits or separate subscription benefit must be checked in the relevant billing console.
Why can the invoice differ from the estimate?
Actual tokens, repeated agent calls, cache hits and writes, reasoning output, tools, regions, taxes and negotiated terms can differ from the entered scenario.