Skip to content
Calcrivo

Total API Cost Calculator

Calculate the full cost of LLM API usage from prompt tokens, completion tokens, model pricing, and request volume.

Inputs

tokens
tokens
requests

Total Cost

$20.00

Cost per Request

$0.020000

Total Prompt Cost

$8.00

Total Completion Cost

$12.00

Step by step

  1. Cost per request: (prompt tokens × input price) + (completion tokens × output price)

    (800 × $0.00001000) + (400 × $0.00003000)

    = $0.020000

  2. Total cost: cost per request × number of requests

    $0.020000 × 1000

    = $20.00

How it works

The total cost of running an LLM in production combines both input (prompt) and output (completion) token pricing, multiplied by request volume: total = (prompt_tokens × input_price + completion_tokens × output_price) × requests. Output tokens are typically priced 2-5x higher than input tokens because generation is more compute-intensive than reading context. Use this to budget monthly or annual LLM spend for a given workload.

Formula

total_cost = (prompt_tokens * input_price + completion_tokens * output_price) * requests

prompt_tokens
Input tokens per request
completion_tokens
Output tokens per request
input_price
Cost per input token (price_per_million / 1e6)
output_price
Cost per output token (price_per_million / 1e6)
requests
Number of API requests

Frequently Asked Questions

Why is completion pricing higher than prompt pricing?

Generating tokens (autoregressive decoding) requires a full forward pass per token, while processing prompt tokens can be batched and parallelized, making input tokens cheaper to serve.

How do I estimate tokens per request?

Use the Token Calculator to estimate tokens from sample prompts and expected response lengths, then plug those averages in here.

Does this include embedding or fine-tuning costs?

No, this covers only standard chat/completion API calls. Embeddings, fine-tuning, and image/audio modalities have separate pricing.

How can I reduce total API cost?

Shorten prompts, cache repeated context, use a smaller/cheaper model where acceptable, and cap completion length with a max_tokens setting.

You might also need