LLM Calc

Working out what an LLM feature will cost usually means opening three vendor pricing pages, squinting at per-million-token rates, and doing arithmetic that is easy to get wrong by a factor of a thousand. These calculators do that arithmetic for you, side by side, with rates that are re-checked every day.

Every tool runs entirely in the browser. Nothing you paste — prompts, text, revenue figures — is uploaded or stored anywhere.

AI & LLM Cost Tools

Developer Tools

Freelance & SaaS Tools

How the pricing data is kept current

Model prices change without much warning, and a calculator built on a rate card someone typed in six months ago is worse than no calculator at all. So the rates here are not typed in by hand. A scheduled job re-reads them every day from the LiteLLM community pricing database and republishes the site when anything moves.

That database is community-maintained. It is not an official vendor feed, and it can be wrong or late. Two things guard against the worst failure mode — a bad number shipping silently:

• Any rate that moves more than 50 percent, or any context window that moves more than 20 percent, is rejected instead of published. A jump that large is far more often a typo upstream than a real price cut, so it gets checked against the vendor's own pricing page by hand before it goes live.
• Every change that does pass is written to the pricing change log, with the date and the old and new values, so you can see what moved and when rather than trusting a page that quietly rewrites itself.

Each tool page carries the date its rates were last synced. Before you commit a budget to a number, confirm it on the vendor's own page — Anthropic, OpenAI, Google.

What these calculators will not tell you

A cost estimate is only as good as the token count behind it, and token counts are where estimates usually go wrong. Three things catch people out repeatedly:

• Input and output are not priced the same. Output tokens typically cost several times more than input tokens. A workload that looks cheap on prompt volume alone can be dominated by the replies.
• The prompt is bigger than you think. System prompts, retrieved context, few-shot examples and prior turns of a conversation are all billed as input, on every single call. In a chat product the conversation history usually outgrows the user's actual question within a few turns.
• Tokens are not words. The rule of thumb that four characters make a token holds for ordinary English prose and breaks badly on code, JSON, non-Latin scripts and long identifiers. The token counter gives a closer estimate than the rule of thumb.

Discounts are not modelled either. Prompt caching, batch endpoints, and negotiated committed-use rates can all move a real bill well below list price. Treat what you get here as a list-price ceiling, not a forecast.

Where to start

Comparing vendors before you build: the LLM API cost calculator puts Claude, GPT and Gemini side by side at the same request volume. Already committed to one vendor: go straight to Claude, GPT or Gemini. Sizing a document pipeline: start with the context window calculator, then the embedding cost calculator. Hitting throughput limits rather than cost limits: the rate limit calculator shows whether requests or tokens per minute is the binding constraint.