Home › FAQ

LLM API Pricing FAQ

This page answers the questions developers ask most about LLM API pricing: how per-token billing works, why output costs more than input, what cached-input pricing and blended prices mean, and how to cut an API bill. Prices referenced are from the October 21, 2026 snapshot of 461 models.

Frequently asked questions

What does 'price per 1M tokens' mean?

Providers bill by tokens (roughly ¾ of a word each). A price of $5 per 1M input tokens means sending 1,000,000 tokens of prompts costs $5. Output (the model's reply) is billed separately and usually costs 3–5× the input rate.

Why is output more expensive than input?

Generating output tokens requires one full model pass per token, while input can be processed in parallel. Across the catalog the median output price is several times the median input price.

What is a blended price?

A single number that mixes input and output at a typical ratio — this site uses 80% input / 20% output. It makes models with very different in/out splits roughly comparable at a glance.

What is cached input (prompt caching)?

If your requests repeat a long prefix (system prompt, documents), providers cache it and bill repeats at the cached-input rate, often 50–90% cheaper. The tables show that rate wherever the provider publishes one.

What is a context window?

The maximum number of tokens a model can consider at once — your prompt plus its reply. Larger windows fit longer documents and conversations but say nothing about answer quality.

Are these the prices I would pay directly at each provider?

Prices come from OpenRouter's public catalog (snapshot October 21, 2026), which mirrors provider list prices. Direct contracts, committed-use discounts and regional pricing can differ — treat these numbers as the list-price baseline and verify before you buy.

How can I reduce my LLM API bill?

The biggest levers: pick a cheaper model for simple tasks (prices span 1000×), cache repeated prompt prefixes, trim input context, and cap output length. The cost calculator shows what each model would cost for your exact volumes.

How often is this site updated?

Daily. Every day a fresh catalog snapshot is taken, prices are re-normalized and all pages are regenerated; the snapshot date is printed on every page.