Input tokens
Everything you send counts: the prompt, system instructions, conversation history, attached documents and tool definitions. Each request is billed for its own input, so a long system prompt is paid for again on every call.
LLM Cost Calculator
PopularEstimate and compare AI API costs based on tokens, usage, and model pricing.
Loading…
How LLM API pricing works
Most language-model APIs charge for the tokens you send and the tokens the model writes back. Here is what goes into the estimate above.
Output tokens
The reply is billed separately and is usually priced several times higher than input. It can't be known in advance, so enter the length you expect. With reasoning models, the “thinking” the model does is typically billed as output too, so include it in your estimate.
Context window
The context window is the most a model can handle in one request: input and output share it. If input plus the expected reply is larger than the window, the request is invalid, so the calculator shows no cost instead of a misleading number.
Tokens per request and monthly usage
Cost per request = input tokens × input price + output tokens × output price (prices are quoted per million tokens). Multiply by how many requests you expect per day or month to plan a budget; the usage calculator does this for each model you selected.
Why different models have different prices
Larger and more capable models cost more to run, and providers price speed, context length and features differently. Tokenizers also differ: the same text can produce more tokens on one model than another, which changes the bill even at the same price per token. Price alone says nothing about which model is right for your task.
Why real billing may differ from an estimate
- Token counts for non-OpenAI models are estimates; providers count with their own tokenizers.
- Cached input, Batch API, priority tiers, long-prompt surcharges, tool or search fees and regional pricing can all change the price.
- Chat APIs add a few tokens of formatting to every message.
- Prices change: confirm on the provider's official pricing page before you commit to a budget.
Related tools: LLM Token Counter, Context Window Checker and JSON Formatter.
About LLM Cost Calculator
Before a feature ships, someone has to answer what it will cost to run at real traffic. Give the calculator a prompt or a token count, say how long the reply will be, and pick the models to price against each other. It shows the cost of one request for each, then scales that by your requests per day or month. Cached input and Batch API rates apply wherever the catalog has a published price for the model.
Features
- Prices up to eight models in one comparison table
- Counted tokens from pasted text or a token count you type
- Cached input share and Batch API rates where the catalog lists them
- Daily, weekly, monthly and yearly projections from request volume
- Warns when expected output exceeds a model's maximum output
- Share link carries your settings, not your prompt
Limitations
- Token counts for non-OpenAI models are estimates, so their costs are estimates too; a token count you type is used as entered.
- Pasted text is limited to 2 MB, and a request whose input plus output does not fit the model's context window gets no cost at all.
- Prices come from a dated catalog at standard rates. Tool, search, regional and priority-tier fees are not included, and models with such rules are marked as a basic estimate.
How it works
- Open the model list, search or filter by provider, and tick the models you want to price.
- Choose Paste text to have the tokens counted, or Enter tokens if you already know the input size.
- Set the expected output tokens for the reply.
- Open Advanced pricing if you want a cached input share or Batch API rates for the models that have them.
- Read the cost estimate and context meter for the focused model, and sort the comparison table by cost, price or provider.
- Enter your requests per day or per month in the usage calculator to see daily, weekly, monthly and yearly cost for each model.
Example input
Input tokens: 2,500 · Output tokens: 500 · Models compared: GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 FlashFrequently asked questions
- Tick one model from each provider in the model list, paste the prompt and set the expected output. The text is counted once per model, with an Exact or Estimated label, and the comparison table puts the estimated costs side by side. Click the Estimated cost header to sort the table by price.
- A cached input share prices that part of the input at the model's cached rate, and Batch API uses its batch rates. Both apply only to models with a published price. They are not combined: with Batch on, the cached share is ignored, and neither is used once the prompt crosses a model's long-prompt threshold.
- Either the request does not fit, meaning input plus expected output is larger than the model's context window, or the catalog has no usable price for it. The calculator shows nothing rather than a misleading number. Shorten the input or the expected output and the cost returns.
- In requests-per-day mode, monthly requests are the daily figure times days per month, which starts at 30 days. Monthly cost is monthly requests times the cost per request, weekly is the daily cost times seven, and yearly is monthly times twelve. By default it reuses your input and output tokens from above.
- No. Tokens are counted by a script in your browser and nothing is posted to Auro Tools. A share link holds the models and numbers only. Your text goes in only if you tick Include my text, up to 2,000 characters, and it sits after the # in the URL, which browsers do not send to servers.
Related tools
- LLM Token Analyzer & Prompt OptimizerCount tokens, estimate API usage, optimize prompts, and discover ways to reduce unnecessary token consumption.
- Context Window CheckerCheck prompt size against a model context window.
- JSON FormatterPretty-print JSON locally.
- Compound Interest CalculatorProject investment growth with compound interest.
- Percentage CalculatorCalculate percentages, ratios, and percentage change.