How AI API pricing works, in plain English
AI models like ChatGPT, Claude, Gemini and Grok charge businesses by the amount of text they read and write. Here is what the numbers on our price table mean and how to estimate a real monthly bill.
What is a token?
A token is a small piece of text, usually part of a word. In English, 1,000 tokens is roughly 750 words. AI companies count tokens to decide what to charge.
Why are input and output priced differently?
Input is the text you send to the model (your question, instructions and any documents). Output is the text the model writes back. Writing costs the provider more work, so output usually costs several times more per token.
What does “per 1M tokens” mean?
Prices are quoted per one million tokens, about 750,000 words. A price of $2 per 1M input tokens means sending one million tokens of text costs $2.
A worked example
Say a small business runs a help chatbot that answers 1,000 customer questions a day, about 30,000 a month. Each question, with its instructions, is about 1,000 tokens going in, and each answer is about 300 tokens coming out. That adds up to 30 million input tokens and 9 million output tokens a month.
The monthly bill is (input tokens in millions × input price) + (output tokens in millions × output price). Here is what that same chatbot would cost on three models from today’s table:
| Model | Input / output $ per 1M | Monthly bill |
|---|---|---|
| GPT-OSS 20B (Groq) | $0.075 / $0.3 | $4.95 |
| Gemini 3.8 Flash (Google) | $0.75 / $3.75 | $56.25 |
| Claude Fable 5.1 (Anthropic) | $10.0 / $50.0 | $750.00 |
Same job, very different bills. A cheaper model is not always good enough, so test a few on your own questions before switching. Our cheapest-for-each-job picks only compare models we judge capable of the job.
How can I lower my AI bill?
Use the cheapest model that does the job well, keep prompts short, ask for shorter answers, and reuse repeated instructions where the provider offers cached-input discounts.
For definitions of tokens, context and batch, see the pricing glossary. For a short checklist, see how to cut your AI bill. See example monthly workloads. Try your own numbers in the savings calculator, or browse prices by model, by provider and in head-to-head compares.
Example figures use standard pay-as-you-go text prices with no batch or cached-input discounts. Prices change, so check the provider’s own pricing page before you buy.