If you have never paid for an AI model directly, the pricing pages can feel like they were written for someone who already knows the answer. They throw around words like "tokens," "context window," and "per million" without ever explaining what any of it means for your wallet. This guide assumes you know none of that.
We are going to build your understanding from the ground up. By the time you finish, you will know what a token is, why AI providers charge the way they do, how to read a pricing page, and how to make a rough estimate of what a project will cost. No prior experience required, and no math more complicated than multiplication.
Think of this as the orientation before the real work. Once these fundamentals click, the more advanced articles in this series will make immediate sense.
What You're Actually Paying For
When you use a hosted AI model — one you access over the internet rather than running on your own computer — you are renting time on someone else's very expensive hardware. AI models run on specialized chips that cost a fortune to buy and operate. Instead of selling you a chip, the provider lets you send requests and charges you for the work each request creates.
The clever part is how they measure that work. They do not charge per request, per minute, or per question. They charge by how much text goes in and how much text comes out. This is fairer than it sounds: a one-word question costs almost nothing, while asking the model to write a 2,000-word report costs meaningfully more, and the pricing reflects that difference automatically.
The One Word You Must Understand: Token
A token is a small piece of text. It is usually about four characters, or roughly three-quarters of an English word. Common short words are one token; longer or unusual words split into several.
Here is the rule of thumb that will serve you well: 750 words is about 1,000 tokens. A page of text is around 500 tokens. A short chatbot reply might be 100 tokens.
Providers quote prices "per million tokens" because individual tokens are so cheap that any smaller unit would be a string of zeros. When you see "$3 per million input tokens," it means you could send roughly 750,000 words of text for three dollars.
Input vs. Output: The Split That Surprises Everyone
This is the single most important beginner concept, and it catches almost everyone off guard.
AI providers charge two different prices in the same request:
- Input tokens — everything you send to the model: your question, any instructions, any background documents.
- Output tokens — everything the model sends back: its answer.
Output almost always costs more than input — typically three to five times more. Why? Generating new text is harder work for the computer than reading existing text. So a request where the model writes a long answer will cost much more than one where it just says "yes" or "no," even if you sent the same question.
The practical lesson for a beginner: long answers are expensive. If you do not need a long answer, ask for a short one.
Reading a Pricing Page Without Fear
Every provider's pricing page has the same bones, even when the layout differs. Look for these four things:
- The model name and which family it belongs to (providers offer several models at different prices).
- The input price per million tokens.
- The output price per million tokens.
- The context window size — the maximum amount of text the model can handle at once.
Everything else on the page is detail you can learn later. If you can find those four numbers, you can estimate a cost. For the full reference on every line item, see our Complete Guide.
Your First Cost Estimate
Let's make this concrete. Suppose you want to build a tool that summarizes customer emails. Here is how a beginner estimates the cost:
- Estimate input. A typical email plus your instructions might be 600 words, so about 800 tokens of input.
- Estimate output. A short summary might be 100 words, so about 130 tokens of output.
- Find the prices. Say the model charges $3 per million input and $15 per million output.
- Do the math. Input: 800 tokens is 0.0008 of a million, times $3 = $0.0024. Output: 130 tokens times $15 per million = $0.00195. Total per email: about half a cent.
- Scale it. If you process 1,000 emails a day, that is about $5 a day, or $150 a month.
That is the entire skill. Once you can do this for one request, you can estimate any workload. For a more detailed walkthrough, our Step-by-Step Approach breaks it down further.
Cheaper Models Exist — Use Them
Providers do not offer just one model. They offer a lineup, like a car manufacturer offering an economy model, a sedan, and a luxury car. The fast, small models can cost a tenth of the flagship and handle most everyday tasks perfectly well.
As a beginner, your instinct will be to grab the biggest, most powerful model. Resist it. Start with a mid-tier or small model, see if the results are good enough, and only move up if you genuinely need to. This one habit will save you more money than any other. Our Best Practices article explains when each tier is worth it.
Frequently Asked Questions
Is using AI models expensive for a small project?
Usually no. For light, experimental use — a few hundred requests a day on a small model — costs are often just a few dollars a month. Costs grow large only when you combine high volume, big models, and long prompts. A beginner's typical project lands at the cheap end.
What's the difference between input and output tokens again?
Input is what you send the model; output is what it generates and sends back. They are billed at separate rates, and output is typically three to five times more expensive because generating text is harder computational work than reading it. Shorter answers cost less.
Do I need a credit card just to learn?
Most providers offer a small amount of free credit when you sign up, which is plenty for learning and small experiments. You will only need to add a payment method once you exceed that free tier or build something with real traffic.
How do I avoid an accidentally huge bill?
Set a spending limit in your provider's billing settings — almost all of them let you cap monthly spend. Start with a small cap. Also use a cheaper model and limit response length while you are learning, so a mistake in your code can't generate thousands of expensive long answers.
What is a context window and do I need to worry about it as a beginner?
The context window is the maximum text a model can read in one request. As a beginner you rarely hit the limit, so you mostly do not need to worry about it. Just know that the more text you stuff in, the more input tokens you pay for on every request.
Key Takeaways
- You pay to rent expensive AI hardware, measured by tokens of text in and out.
- A token is about four characters; 750 words is roughly 1,000 tokens.
- Input and output are billed separately, and output costs three to five times more.
- To estimate cost: count input and output tokens, multiply by the prices, scale by volume.
- Start with a small or mid-tier model and set a billing cap before you experiment.
- The fundamentals here unlock every other article in this series.