LLM cost calculator

How the math works

Every provider bills per token, with separate prices for what you send and what comes back:

cost per call = input tokens × input price
              + output tokens × output price
              − cached input tokens × (input price − cache-read price)

Prices are quoted per million tokens. For example, a call to Claude Sonnet 5.5 with 3,000 input and 500 output tokens costs 3,000 × $2.00/1M + 500 × $10.00/1M. Multiply by calls per day and by 30 for the month.

• A token is about four characters of English, or three quarters of a word. Code, JSON and other languages take more.
• Output costs more than input, usually 3–8×, because it is generated one token at a time. Long answers, reasoning tokens and verbose JSON add up fast.
• Cached input is cheap. When the start of your prompt repeats, most providers bill it at 10–50% of the input price. See how prompt caching is priced.
• Reasoning tokens are output tokens. A model that thinks for 2,000 tokens before a 200-token answer bills 2,200 output tokens.

Where to get your real numbers

The calculator is only as good as its inputs. Your provider’s usage page shows tokens by model and by key, but not by feature, so “what does the summarizer cost?” stays a guess. Two ways to get it:

1. Log usage.prompt_tokens, usage.completion_tokens and cached tokens from every response, tagged with the feature that made the call.
2. Or wrap your client with Gin in one line, fetch: gin({ useCase: 'summarize' }), and get spend by use case in a free daily email, with cheaper models already graded on your calls.