AI Cost Calculator
Estimate your AI API costs before you build. Supports GPT-4o, Claude, Gemini, and more. Runs on your device — nothing is uploaded, nothing is stored.
The AI Cost Calculator turns token counts into dollars. Choose a model from the list, enter how many input tokens and output tokens one request uses, and say how many requests you will make. It shows the input cost, the output cost and the total for a single request, and when you enter more than one request it adds a total for the whole batch.
The model's input and output prices per million tokens are shown right under the selector, so you can see exactly what the arithmetic uses. Each cost is the token count divided by one million and multiplied by that price, calculated separately for input and output. The list covers OpenAI, Anthropic, Google, Meta (Llama, via third-party hosts), Mistral and DeepSeek models, and the starting values are 1,000 input tokens, 500 output tokens and one request on GPT-4o. Use Reset to go back to them.
Why this runs on your device
You are typing numbers, not text, so there is no privacy stake to claim. The benefit of running in the page is that the cost updates as you type, you need no API key, and you can try different models and volumes in seconds.
- Choose a model from the list and check its input and output prices.
- Enter the input tokens, output tokens and number of requests.
- Read the input, output and total cost per request, plus the total for all requests.
- It estimates from list prices only. Cached-input discounts, batch pricing, long-context surcharges, fine-tuned model rates and charges for images or audio are not modelled.
- The prices live in a table stored in the site, so they are only as current as its last update. Providers change prices without notice, and this page will not detect that; confirm the rate on the provider's pricing page.
- You supply the token counts. The calculator does not read your prompt, and it only counts what you enter: system prompts, tool definitions, conversation history sent back each turn and a model's hidden reasoning, if it bills for it, are all things you must add yourself.
- Token counts are whole numbers: decimals are cut off. A blank or zero request count is treated as one request, and results are shown in US dollars.