Type how much text your app or agent sends and receives in a month, and see what it costs on 18 AI models, at each provider's own list price. Prices were read off the providers' pricing pages on 19 November 2026: standard rate, no batch discount, no cache hits. One token is about four characters of English, so 1,000 tokens is roughly 750 words (what a token is).
How it works
Cost = input tokens × the input price + output tokens × the output price, each per million tokens. Words are converted at 4 characters per token (Perplexity's guide) and about 5 characters per word, so 1 million words is about 1.33 million tokens; for other languages and code the real count is usually higher. Reasoning models also bill their hidden thinking as output, so measure a few real runs before trusting any estimate, this one included.
What it leaves out
Discounts and surcharges that can move a real bill a long way: cached input (90% or more off at most providers), batch jobs (half price at OpenAI, Anthropic, Google and Mistral), DeepSeek's off-peak half price, long-prompt surcharges and regional endpoints. Each is set out on What do AI model APIs cost?. Google's Gemini 4 Carbon and Argon double to $4 / $20 per million tokens on 1 December 2026 (all dated price changes).
The page this is about: What does an AI model API really cost to run? — every figure sourced to the vendor's own page, with the date it was checked.