Golden Prompt Lab

Token Economics

AI isn't free to run. Behind every response is a data center drawing electricity, and the pricing model reflects that. Understanding how tokens translate to dollars helps you make smarter decisions about which AI to use and how to use it.

How pricing works

Providers charge separately for input tokens (what you send) and output tokens (what the AI generates). Output tokens are almost always more expensive — generating text requires more compute than reading it.

Prices are quoted per million tokens (written as $/1M). That sounds like a lot, but consider: a typical back-and-forth conversation might use 500–2,000 tokens. A million tokens is roughly 750,000 words — the length of several novels.

The spectrum of models

Not all models cost the same because not all models are equally capable — or equally large. The general pattern:

Model tierSpeedCostBest for
Small / fast (e.g., Haiku, GPT-4o mini)Very fastVery lowSimple tasks, high volume
Mid-tier (e.g., Sonnet, GPT-4o)FastModerateMost everyday tasks
Large / flagship (e.g., Opus)SlowerHighComplex reasoning, long docs

The hidden cost of verbosity

One practical takeaway: the longer your prompt, the more you pay. Repeating context, being overly polite ("Please, if you don't mind..."), or pasting in large documents all add tokens. This matters more for developers building applications than for casual users — but it shapes how AI products are designed.

What about free tiers?

Apps like ChatGPT and Claude.ai offer free access because the companies subsidize it to grow their user base. The underlying API still costs money — the companies are just paying it for you up to a point. When you hit rate limits or get slower responses, you're bumping into those economics.

Gear up

The Tokenizer Explorer lab below lets you type anything and see the cost math unfold in real time — across multiple models, side by side.