LLM API Cost Calculator
A GPT-4.1 request with 2,000 input and 500 output tokens costs $0.008, so 1,000 requests a day come to $240.00 a month at the prices checked on 2026-10-15.
How to use
- Pick a model, then enter the input and output tokens of one request and how many requests you send per day or per month.
- Turn on the prompt cache and set the share of input tokens that hit the cache, or turn on the batch discount. Options a model does not offer are ignored and flagged.
- Read the monthly cost, then the cost per request, per day and per year. The table below prices every model for the same usage, cheapest first.
How it works
Cost per request = (input tokens × input price + output tokens × output price) ÷ 1,000,000. Prices are US dollars per million tokens.
With the prompt cache, the input price is a weighted average: hit rate × cached input price + (1 − hit rate) × normal input price.
The batch discount swaps in the vendor batch prices, which are 50% of standard on the pages checked. Where a vendor lists a cached price for batch it is used, otherwise the standard cached price.
Daily cost = cost per request × requests per day. A month is counted as 30 days, so monthly = daily × 30 and yearly = monthly × 12.
Not included: cache write surcharges, cache storage fees, long-prompt price tiers, tool and search fees, taxes. Prices are copied from the official pages on the date shown and are not converted to other currencies.
FAQ
Where do the prices come from?
From the official pricing pages of OpenAI, Anthropic, Google, Mistral and DeepSeek, linked under the calculator, on the date shown as the check date. A value that could not be confirmed is left out rather than guessed.
How do I count tokens?
Use the usage numbers returned by the API, or a vendor tokenizer. As a rough guide, one token is about four English characters or three quarters of a word. Other languages usually take more tokens per character.
What does the cache hit rate mean?
It is the share of your input tokens read from the prompt cache instead of processed fresh. A long system prompt that repeats in every request can give a high rate. Cache hits cost much less than normal input.
Why is a model marked with a dagger in the table?
It does not offer the cache or batch option you selected in the figures used here, so its row is priced at the full standard rate.
Why is the real bill different?
Reasoning tokens, tool calls, long prompts above a price tier, cache writes and storage, taxes and rate changes are not modelled. Treat the result as an estimate and check the official page before you decide.
Why are amounts in US dollars only?
All five vendors list prices in dollars. Converting with an exchange rate would add a number this page cannot verify, so the amount is shown as dollars in your local number format.