API Pricing

API pricing is the price developers pay to call an LLM from their programs. Most developers of models state prices in US dollars per million tokens. The amount the model reads (input) and the amount it writes (output) are counted separately in tokens, and the bill is each count multiplied by its price per token, added together. The output price is usually higher than the input price.

For models that think before they answer, the tokens spent thinking before the answer are also billed as output. Google's pricing page states that its output prices include thinking tokens. Even for the same question, the higher the thinking setting, the higher the price.

On the pages for choosing by use, the standard price is a blend of input 3 : output 1. This follows a view common in the industry that models read more than they write. For having the model read documents and answer, and for classification and extraction, reading makes up most of the tokens, so these uses compare by the input price.

The price of one request can be estimated from the number of tokens. By this site's estimate, an ordinary question takes about 1,000 tokens for the question and answer together. For a model at $1 per million tokens, one request of 1,000 tokens costs about $0.001. Yen prices on the Japanese pages are estimates converted from US dollars with the European Central Bank's reference rates. The European Central Bank itself says these rates are for information only and strongly discourages using them for transactions, so they differ slightly from what you actually pay.

Paid plans for chat apps and API pricing are separate systems. According to Anthropic, paid app plans and the developer-facing API are separate products. API usage is billed separately (some plans include monthly API credits). If you only use the app, you do not need to pay API prices.

Developers of models sometimes change their prices, and some ways of using the API cost less. As of October 2026, Anthropic gives a 50% discount on input and output for the Batch API, which processes requests later in bulk. Before using a model, check the prices on its developer's pricing page.