LLM Token Cost Calculator

It uses the standard per-million-token pricing that OpenAI, Anthropic, Google, and most LLM APIs publish. Input cost is input tokens divided by 1,000,000 times your input price, output cost is output tokens divided by 1,000,000 times your output price, and the two add up to the cost per request. That per-request figure is then multiplied by your requests per day, and the daily total is multiplied by 30 for a monthly run rate.

Almost every provider charges more for output (generated) tokens than for input (prompt) tokens, often two to four times as much. Keeping the two rates apart lets you enter the exact prices from your provider's pricing page and see which side of the request drives your bill. The result panel also shows the input share so you can tell at a glance whether prompts or responses dominate.

A token is a chunk of text the model reads or writes, usually a short word or part of a word. As a rough rule of thumb, 1,000 tokens is about 750 English words, so a token averages around four characters. Counts vary by model and language, so for precise billing use your provider's tokenizer or the token usage reported in the API response.

Yes. The calculator does not assume a specific model. You type in the input price and output price per million tokens for whatever model you are using, so it works for GPT, Claude, Gemini, open-weight models on a hosted endpoint, or anything else billed per token. Just copy the two prices from the provider's pricing page.

This is a list-price estimate, not a billing guarantee. It does not account for cached or batch-discounted prompts, free tiers, image or audio tokens, fine-tuning surcharges, tiered enterprise rates, or taxes, and it assumes a flat 30-day month. Treat the output as a planning figure and confirm the exact numbers on your provider's invoice.

Set requests per day to 1. The cost per request value then equals your single-call cost, and the daily and monthly figures simply repeat that call once per day over 30 days. To compare two models on the same workload, keep the token counts fixed and swap in each model's input and output prices.