What Are Tokens and Why Do They Cost Money?

Tokens: The Currency of AI

If you're building with large language models, you'll quickly encounter the word "token." Understanding tokens is essential because every API call you make is priced in tokens — and misunderstanding them can lead to surprise bills.

What Is a Token?

A token is a chunk of text that an LLM processes. It's not exactly a word and not exactly a character — it's somewhere in between. As a rough rule of thumb:

• 1 token ≈ 4 characters in English
• 1 token ≈ ¾ of a word
• 100 tokens ≈ 75 words

The word "hamburger" is 3 tokens: "ham", "bur", "ger". Common words like "the" or "is" are single tokens. Numbers are often tokenized digit-by-digit.

Input vs. Output Tokens

Every API call has two token counts:

• Input tokens: Everything you send to the model — your system prompt, user message, context, examples
• Output tokens: Everything the model generates in response

Crucially, output tokens are more expensive than input tokens, typically 2-4× more. This is because generating each output token requires a full forward pass through the model, while input tokens can be processed in parallel.

Why Do Prices Vary So Much?

Model pricing spans a 100× range. Here's why:

Model Size

Larger models (more parameters) require more compute per token. GPT-5 with hundreds of billions of parameters costs more than GPT-5 Nano with far fewer.

Infrastructure

Running inference on cutting-edge GPU clusters (H100s, B200s) is expensive. Providers price to cover hardware costs, power, and cooling.

Demand & Competition

Pricing is also strategic. Providers sometimes price aggressively to gain market share, or premium-price flagship models where they have a quality advantage.

Context Windows: Bigger Isn't Always Better

Modern models advertise massive context windows — 200K, even 1M tokens. But here's the catch: you pay for every token in the context, whether the model uses it or not.

Stuffing 100K tokens of context into every request when you only need 2K is like leaving all the lights on in a skyscraper. It works, but it's wasteful.

How to Estimate Your Costs

1. Count your tokens: Use a tokenizer tool or estimate at 1 token per 4 characters

2. Separate input and output: Track how many tokens you send vs. receive

3. Multiply by price: Apply the per-million-token rate for your chosen model

4. Multiply by volume: Estimate your monthly request count

Or just use our Calculator — it does all of this for you automatically.

Related Reading

• GPT-5 vs Claude 4 Pricing — See real token prices compared side by side for the two leading model families.
• Are Open-Source Models Really Cheaper? — The hidden costs of self-hosting vs. managed APIs.
• 5 Tactics That Cut Our LLM Costs by 50% — Once you understand tokens, here's how to spend less on them.
• Subscription vs. API — Compare $20/month subscriptions against pay-per-token API access.