What Are Tokens in AI? A Plain-English Explainer

A token is the basic unit of text an AI model reads and generates – a word, part of a word, or a punctuation mark. Every prompt you send and every response an AI model gives back gets broken down into these pieces first. Understanding tokens matters beyond the technical layer, too: most AI vendors now price their tools by token, which is part of why AI-related software costs can shift from one month to the next.

Quick Answer

A token is the basic unit of text that AI language models process – smaller than a sentence, and often smaller than a word.

• A token can be a whole word ("the"), part of a word ("token" + "ization"), or a single punctuation mark.
• As a rough rule of thumb, 1 token works out to about 4 characters or roughly 0.75 words of English text, per OpenAI's own documentation.
• Every AI request has input tokens (what you send) and output tokens (what the model generates back) — and output is typically priced higher.
• A context window is the maximum number of tokens a model can take in and respond to at once.
• Bottom line: understanding tokens is the foundation for understanding what you're actually paying for in AI-powered software.

Key Takeaways for Buyers

• Every prompt sent to an AI model, and every response it generates, is measured – and increasingly billed – in tokens, not in words or seats.
• Most AI vendors now price by token or by a token-based "credit," and vendors define "credit" inconsistently. This is a major reason AI-related software costs can shift month to month and be difficult for finance teams to forecast and track.
• Tokens are the reason AI bills feel unpredictable compared to traditional per-seat software: usage varies by content, not by headcount.

What Is a Token, Exactly?

A token is the smallest chunk of text an AI model can read or write – not quite a word, not quite a letter, but somewhere in between. According to OpenAI's own documentation, the string "ChatGPT is great!" breaks down into six tokens: "Chat," "G," "PT," " is," " great," and "!." Notice that a single word ("ChatGPT") can split into multiple tokens, while short, common words often stay whole.

Here's how tokens compare to the units most people think in instead:

Because tokens don't map cleanly to words, a 500-word document and a 500-word email can consume noticeably different numbers of tokens depending on vocabulary, punctuation, and formatting.

How Does Tokenization Work?

Before an AI model can process any text, it has to convert that text into tokens – a step called tokenization. Most modern models use a method called “subword tokenization”, which keeps common words as single tokens but breaks rare, technical, or made-up words into smaller pieces the model has seen before.

This is why a plain, everyday sentence tends to use fewer tokens than a sentence packed with brand names, jargon, or unusual terms. The model isn't guessing at whole words; it's assembling meaning from a fixed vocabulary of token pieces, the same way it was trained to.

Why Tokens Matter for AI Cost and Performance

Tokens aren't just a technical detail for IT leaders and engineers – they're increasingly the unit AI vendors use to bill you. Nearly every commercial AI API or tool prices based on some form of token volume, and that has two direct effects on cost.

1. Input and output tokens are usually priced differently, with output tokens typically costing more than input tokens. 
2. Every model has a context window – a cap on how many tokens it can take in and respond to in a single exchange – and that cap shapes how much information you can feed a model at once.

Context windows have grown quickly and vary by model. As of July 2026, most current flagship models support context windows around 1 million tokens:

These figures shift every few months as vendors ship new model versions, so treat them as a snapshot rather than a fixed spec. This is worth confirming against the vendor's current documentation before relying on a specific number.

This is also where token-based pricing gets confusing for finance and procurement teams. According to Tropic's analysis, credit-based AI pricing grew 126% year-over-year in 2025, and vendors define what a "credit" actually represents inconsistently – sometimes a credit maps to one token, sometimes to an API call, sometimes to a broader "agent action." Tropic's data also shows AI-driven renewal increases running 20–37% higher than historical software uplifts, a gap partly driven by this same lack of standardization in how usage gets measured and billed.

Tokens vs. AI Credits: Are They the Same Thing?

Tokens and credits are sometimes the same thing, but not always. It depends entirely on how a specific vendor defines a credit, and that definition isn't standardized across the industry.

Some vendors set up their pricing so that one credit equals one token processed. Others define a credit as one API call, or one "agent action," regardless of how many tokens that action actually consumes under the hood. Tropic's own usage-based pricing research found that this inconsistency is exactly what makes forecasting AI spend difficult for buyers: two vendors can both sell "credits," yet one credit from Vendor A might cost you far more in actual token usage than one credit from Vendor B.

Where Tokens Fit Into Your AI Spend

Now that you know what a token is, the natural next question is how tokens turn into what you're actually being charged. That's a bigger topic than this page covers, since it involves how vendors structure pricing tiers, how usage scales across a growing tech stack, and what levers buyers have to negotiate.

For a category-by-category breakdown of where AI costs come from and how to manage them, see Tropic's guide on managing AI costs. If your team wants a closer look at what's actually happening with AI consumption across your software portfolio, get a demo to see how Tropic's spend intelligence surfaces this for finance and procurement teams.

FAQ: Tokens in AI

What is a token in AI, in simple terms? 

A token is the basic unit of text an AI model reads and generates – a word, part of a word, or a punctuation mark. Models process everything in tokens, not in whole words or sentences, before turning it back into readable text.

How are tokens different from words? 

Words are how people naturally split language; tokens are how AI models split it internally. A single word can be one token or several, depending on how common or unusual it is. As a rough estimate, 1 token equals about 0.75 words of English text.

Why do AI tools and models charge by token? 

Token volume closely tracks the actual compute an AI model uses to process a request, so vendors bill by token, or by a token-based "credit," to align cost with usage. This has replaced flat, per-seat pricing for many AI-powered tools and APIs.

What is the difference between input tokens and output tokens? 

Input tokens are the text you send to an AI model; output tokens are the text it generates back. Most vendors price these separately, and output tokens are typically priced higher than input tokens.

What is a context window, and how does it relate to tokens? 

A context window is the maximum number of tokens a model can take in and respond to in a single exchange, combining both your input and its output. As of July 2026, most current flagship models – Claude Sonnet 5, GPT-5.5, Gemini 3.1 Pro – support context windows around 1 million tokens, though this changes as new versions ship.

Are AI tokens the same thing as AI credits? 

Sometimes, but it depends on the vendor. Some vendors define one credit as one token; others define it as one API call or one agent action. There's no industry-wide standard, so the same number of credits can represent very different amounts of actual usage across vendors.

Why does my AI bill change even when my usage feels the same? 

Because token consumption depends on the specific content processed, not just how often you use a tool. Longer or more complex prompts and responses use more tokens, and vendors can also change credit definitions or pricing at renewal, which can shift your bill even when your day-to-day usage feels consistent.