Prices come per million tokens

Most companies that sell access to an AI model post a price sheet. The numbers look tiny because they are quoted per million tokens. A token is a small piece of text. Hello, AI puts a 200-word answer at roughly 260 tokens, so a token is about three quarters of a word. A million tokens is then around 750,000 words.

Take Claude Haiku 5.5, a small model Anthropic released on October 7, 2026. For most prompts it costs $0.10 per million input tokens and $0.50 per million output tokens. Input is everything you send: your question, any instructions, any document you paste in. Output is what the model writes back.

Why the reply costs more

Output usually costs more than input. For Haiku 5.5 it is five times as much. Back in March 2023, GPT-4 charged $30 per million input tokens and $60 for output, twice as much.

The reason is in how a model works. When it reads your prompt, the chips can take in many tokens in one pass. When it writes, it makes one token, adds it to the text, and runs the whole model again for the next one. Each new token means reading all of the model’s weights out of memory once more.

Reading is a crowd walking through a wide door together. Writing is a queue where each person is stamped in turn. So a short question with a long answer can cost more than a long question with a short answer.

Long prompts can cost more per token

Some price sheets have tiers. Haiku 5.5 keeps its low price for prompts up to 100,000 tokens. Above that, input rises to $0.50 per million and output to $2.50. Haiku 4.5, the model before it, cost $1 and $5. Anthropic says that makes Haiku 5.5 about 90% cheaper for shorter requests, 50% cheaper for longer ones, and around 75% cheaper on average.

Same text, more tokens

Each model has a tokenizer, the tool that cuts text into tokens, and two tokenizers can cut the same text into different numbers of pieces. Haiku 5.5 uses a newer one. Anthropic’s documentation says the same text counts as about 30% more tokens than on Haiku 4.5.

A document that was 1,000 tokens on Haiku 4.5 becomes about 1,300 on Haiku 5.5. The input price fell from $1 to $0.10, a tenth. So that document’s input bill falls by more than four fifths, a smaller drop than the price sheet alone suggests. When you compare two models, compare what the same text costs on each.

In one sentence: you pay per million tokens, writing costs more than reading, and the real bill depends on how many tokens your text becomes.