Free tool

AI token cost calculator

Enter the tokens in a typical request and how many requests you send a month. The table prices that job on 116 text and embedding models, at each maker's list price in US$.

Estimate input tokens from text

In English, about 4 characters make one token. Code, JSON and most other languages use more tokens per character, so treat this as a rough estimate.

Text models

Cost of the job above on each model, lowest first
ModelProviderInput / 1MOutput / 1MPer requestPer monthPer month on Aight
The cost table needs JavaScript. Prices per model are on the models page.

How the calculator works

Cost per request is input tokens times the input price, plus output tokens times the output price, divided by 1,000,000. Cached input tokens use the model's cached-input price where it has one, and the full input price where it doesn't. When a prompt is longer than a model's long-context threshold, that model's higher long-context prices apply. A model whose context window is smaller than your prompt is left out.

Reasoning models bill their thinking tokens as output, so add them to output tokens. Embedding models have no output price. Image, speech and per-image models are priced differently and aren't in this table.

Where the prices come from

Each price is the US$ list price per 1M tokens on Aight's models page, catalog dated 18 Nov 2026. Aight charges the maker's list price on most models; a few hosted models are priced at the cloud host's rate. Check the maker's own pricing page before you commit to a large job.

The last column

From outside India, a $100 recharge costs $103 with 3% processing and adds $110 of balance, because every recharge gets 10% extra. Each $1 of list-price usage then costs about $0.94, and the last column applies that. India accounts recharge in ₹ with 18% GST, so the column doesn't apply to them. The details are in the refund policy.

Questions

How many tokens is a word?

In English, about 0.75 words per token, so 1,000 tokens is roughly 750 words. Each model family has its own tokenizer, so the same text can come to a different count on Claude, GPT and Gemini.

Where do real token counts come from?

Every API response reports them: usage.prompt_tokens and usage.completion_tokens in Chat Completions. Log them for a day of real traffic and put the averages here.

Why can a bill come in higher than the estimate?

The usual causes are long system prompts and chat history resent on every turn, reasoning tokens, retries, and tool calls that add extra requests. Logging usage per request shows which one it is.