# What is an AI token? Definition and business implications

> A token is the elementary unit of text that an AI model manipulates, generally a word fragment of 2 to 4 characters, depending on the model and the language. It is both the model's unit of computation and the AI providers' unit of billing, counted separately on input and output.

## Definition
To split text into tokens, models use an algorithm (generally *Byte Pair Encoding*, BPE) that assigns a unique token to the most frequent character sequences in their training corpus. Consequence: a common word like “hello” often fits in a single token, whereas a rare term like “disintermediation” will be split into several pieces. On the operational side, each generated token requires a complete pass through the model's neural network, billions of mathematical operations. Hence a dual cost: energy (electricity consumed by servers) and economic (billed by providers). Anthropic prices as of October 2026 illustrate the orders of magnitude, per million input and output tokens: Claude Haiku 5.5 at $0.10/$0.50 (prompts up to 100,000 tokens), Sonnet 5.5 at $2/$10, Opus 5.5 at $4/$20, Fable 5.1 at $10/$50. The output/input ratio is 5:1 across the whole line-up. Be careful with comparisons: according to Anthropic, recent Claude models split the same text into about 30% more tokens than earlier ones.

## Example
A 50-employee SME uses an AI assistant via the Claude Sonnet 5.5 API to draft its commercial responses. A typical exchange consumes 500 input tokens and 800 output tokens, about $0.009 at September 2026 prices. Over 1,000 monthly exchanges, the bill remains anecdotal ($9). But double the length of prompts and responses through lack of method, and it rises to $18. At a mid-cap with 50,000 monthly exchanges, the gap becomes structural: $450 vs $900, about $5,400 annual difference for the same business usage.

## For a leader
The token is the unit that transforms AI from an abstract topic into a measurable budget line. Three implications for the executive. First, the total cost of ownership of your AI solutions is calculated in tokens, not in flat subscriptions. Industrialised use without discipline inflates the bill with no quality gain: doubling the length of exchanges doubles the cost. Second, prompting sobriety (precise instructions, calibrated contexts, no unnecessary verbosity) is not a technical nicety but a direct economic lever. The 5:1 output/input ratio implies that outputs weigh more in the bill than inputs: mastering the length of generated responses is a priority. Third, environmental footprint follows the same logic. Every token has a real energy cost; prompting discipline simultaneously reduces the financial bill and the carbon impact.

## FAQ

**What exactly is a token?**
It is the elementary unit of text an AI model works with, usually a word fragment of about 3 or 4 characters. A common word like “hello” counts as a single token, while a rare term is split into several pieces. It is at once the model's unit of computation and the providers' unit of billing.

**Why does a token belong on the budget rather than in the IT department?**
Because the total cost of ownership of your AI solutions is counted in tokens, not in flat subscriptions. Providers bill input and output separately, and at Anthropic's October 2026 prices the output-to-input ratio is 5:1 across the whole line-up. Undisciplined use inflates the bill with no gain in quality: doubling the length of exchanges doubles the cost.

**How do you keep the token bill down?**
Through prompting discipline: precise instructions, calibrated context, no needless verbosity. Since outputs weigh five times more than inputs in the bill, controlling the length of generated responses matters most. The same discipline lowers both the financial cost and the energy footprint, because every token requires a full pass through the model's neural network.

**What cost gap does this create at company scale?**
For an SME running 1,000 monthly exchanges on Claude Sonnet 5.5, doubling the length of prompts and answers through sloppy method takes the bill from $9 to $18, trivial amounts. At a mid-cap with 50,000 monthly exchanges, the same gap becomes structural: roughly $450 versus $900, about $5,400 a year for identical business use.

**Can you compare token counts directly from one model to another?**
Not safely. According to Anthropic, recent Claude models split the same text into about 30% more tokens than earlier ones. Judging two offers on price per million tokens alone, without accounting for how each model slices the text, distorts the comparison.

## See also

- [LLM (Large Language Model)](https://wyp.agency/en/glossary/llm/)
- [Prompt](https://wyp.agency/en/glossary/prompt/)
- [Context window](https://wyp.agency/en/glossary/context-window/)
- [Inference](https://wyp.agency/en/glossary/inference/)
- [Cost per token](https://wyp.agency/en/glossary/cost-per-token/)

## Sources

- Anthropic documentation, Pricing: Claude Haiku 5.5, Sonnet 5.5, Opus 5.5, Fable 5.1 rates and tokenizer note, October 2026 (https://platform.claude.com/docs/en/about-claude/pricing)
- Neural Machine Translation of Rare Words with Subword Units, Sennrich, Haddow & Birch, ACL 2016 (BPE algorithm) (https://arxiv.org/abs/1508.07909)
- Models overview, Claude platform documentation, Anthropic (https://platform.claude.com/docs/en/about-claude/models/overview)

---
Entry from the WYP.agency AI glossary · https://wyp.agency/en/glossary/token/ · Last updated : 2026-10-14