LLM API cost calculator for 340+ models

Every major LLM API bills by the token: small chunks of text that average about four characters in English. Input tokens (everything you send: system prompt, conversation history, retrieved documents) and output tokens (the model’s reply) have separate prices, quoted in US dollars per million tokens. This calculator applies those published prices to your workload.

The formula

For one request, the cost is:

(uncached input × input price + cached input × cached price + output × output price) ÷ 1,000,000

The daily cost multiplies that by your requests per day. The monthly figure uses the average month length of 365 ÷ 12 = 30.4 days, and the yearly figure uses 365 days.

Output tokens are the expensive part

Output is priced at 3× to 5× the input price for the popular models below, because each output token is generated one after another. Two things inflate output more than people expect. First, reasoning models bill their hidden “thinking” as output tokens, so a short visible answer can still cost thousands of tokens. Second, a verbose system prompt (“explain your reasoning step by step”) makes every reply longer. Asking for concise answers is often the cheapest optimisation available.

Prompt caching

If many requests begin with the same text, providers can keep that prefix in a cache and charge 2% to 27% of the normal input price to read it again. The calculator assumes the cache is already warm. Some providers also charge a one-time premium to write to the cache, which matters if your prefix changes often. Caching only applies to an identical prefix, so put the fixed parts of your prompt first and the parts that change last.

Batch API and long-context pricing

Batch APIs run requests in the background, usually within 24 hours, at roughly half price. When you turn on batch mode, the calculator uses each provider’s published batch price, and it tells you if a model has none. Some models also switch to a higher rate once a prompt passes a size threshold, such as 200,000 tokens. When that happens the higher rate covers the whole request, and the calculator shows a note.

Choosing a cheaper model

The alternatives table ranks every model whose context window and output limit can handle your request, from cheapest up. It doesn’t judge quality: the cheapest model may not be good enough for your task. Use it to shortlist candidates, then test them on real prompts before switching.

Prices come from the OpenRouter models API and the open-source LiteLLM price list and are refreshed daily (last updated 2026-10-09). For open-weight models such as Llama, the price is a typical hosted price; running them on your own hardware costs something else entirely. The methodology page explains how the data is collected and checked.