AI API Cost Calculator

What Is the AI API Cost Calculator?

The AI API Cost Calculator is a free online tool that answers one practical question: how much will this AI request cost me? You choose a provider and a model, enter how many tokens your request uses, and the tool returns the price in US dollars for a single request and for a whole month of traffic at your chosen volume.

It is built on the public pricing catalog at models.dev, so the rates come from the providers themselves rather than from guesswork. The catalog covers thousands of models from OpenAI, Anthropic, Google, DeepSeek, Meta, Mistral, Alibaba, xAI, Amazon, Cohere and many others. Pricing is stated the way providers bill it: in USD per 1 million tokens, with separate rates for input, output, cached text, reasoning steps and audio.

A token is the billing unit of every large language model. As a rough guide, one token is about four characters of English text, so a 400-word article is close to 500 tokens. Because input and output are priced differently, and because cache and reasoning have their own rates, a mental calculation is almost always wrong. This tool does it exactly.

Why Use It — and Who It Is For

Model pricing is easy to misread. A rate that looks tiny per thousand tokens becomes a serious bill at production volume, and an apparently expensive model can turn out to be the cheapest once prompt caching is counted. Real people use this calculator for:

• Developers who need a cost answer before wiring a model into a product.
• Founders and product managers who are budgeting a monthly AI bill or choosing between providers.
• Consultants and agencies who must quote a fixed price for a project with unknown traffic.
• Students and researchers comparing open-weight models for an experiment on a tight budget.
• Anyone with an unexpected invoice who wants to see where the money went, component by component.

The question people actually need answered is rarely "which model is smartest". It is "which model fits my budget at ten thousand requests a day", and that is what this calculator measures.

How This Tool Works

Everything on the page comes from one pricing catalog. When the page loads, the browser asks our server for the provider list. The server downloads the official models.dev API file, caches it for 24 hours, and also keeps a stored copy on disk, so the tool keeps working even if the upstream site is briefly unreachable. Every response carries the last update time and a stale flag, and the page shows a warning when pricing data could not be loaded at all.

When you change a number, the page waits 350 milliseconds before sending the request, so fast typing does not flood the server. The calculate endpoint then runs these steps on our backend:

1. Lookup. Your provider and model are matched against the catalog. If either is unknown you get the message "Model not found in catalog."
2. Validation. At least one token count must be greater than zero, otherwise the server replies "Enter at least one token count greater than zero."
3. Clamping. Each token field is clamped to a maximum of 50,000,000 tokens, and requests per month is clamped to a range of 1 to 100,000,000, so a typo cannot produce a nonsense result.
4. Line items. Every component with tokens above zero and a published rate becomes one line: tokens divided by 1,000,000, multiplied by the rate. Components with no tokens, or with no published rate, are skipped rather than guessed.
5. Totals. Line items are summed into a per-request cost, then multiplied by your monthly request count to produce the monthly projection.
6. Warnings. The server checks your input side (input plus cache read, cache write, reasoning and audio input) against the model context window, and your output tokens against the model maximum output limit, and warns you when either is exceeded.

Rates are applied with the same fallbacks providers use. Cache write falls back to cache read and then to the input rate when a provider does not publish it. Reasoning tokens use the reasoning rate when available and otherwise the input rate. Audio input and audio output fall back to the standard input and output rates.

Key Features

• Three working tabs. Calculate Cost, Compare Models and Pricing Browser, each with its own panel.
• Live pricing. Rates are refreshed from models.dev on a 24-hour cycle, and every response reports the last update time plus a stale flag.
• Full token breakdown. Input, output, cache read, cache write, reasoning, audio input and audio output are each billed at their own rate and listed in a table with tokens, rate and cost.
• Four headline numbers. Per request, weighted price per 1M tokens, monthly total and tokens per request.
• Monthly projection. A slider from 1 to 10,000,000 requests per month recalculates the bill as you drag it.
• Quick scenario chips. Chat assistant, RAG pipeline, Code generation, Deep reasoning and Voice assistant fill the token fields with realistic starting values in one click.
• Side-by-side comparison. Put 2 to 4 models against one shared usage profile and get a table plus a monthly-cost bar chart.
• Pricing browser. Search the catalog, sort by input, output or cache-read price, switch between cheapest first and priciest first, filter to reasoning models only, and load any row into the calculator with its Use button.
• Token estimator. Paste any text for a quick token count using the four-characters-per-token rule of thumb, which suits English and over-estimates for Chinese or Japanese.
• Context and output warnings. Automatic alerts when your numbers cannot fit the model you selected.
• Model meta badges. Provider, status, reasoning, tool calls, open weights, context size and maximum output are shown next to the model you pick, with deprecated models marked in red.

How to Use — Step by Step

1. Open the Calculate Cost tab, which is selected by default.
2. Choose a Provider from the dropdown. The provider list loads from the live catalog.
3. Choose a Model. Type in the filter box above the list to narrow it, then click the model you want. Meta badges appear underneath.
4. Enter your usage per request: Input tokens (default 1000), Output tokens (default 500), and any of Cache read, Cache write, Reasoning, Audio input and Audio output that apply to your workload.
5. Drag the Requests per month slider to your real traffic volume, or click a quick scenario chip such as RAG pipeline to fill the fields with sensible defaults.
6. Read the results: four stat cards, then the cost breakdown table, then the note summarizing per-request cost and monthly total. Warnings appear above the note if your input exceeds the context window or your output exceeds the model limit.
7. Optional: paste your prompt into the Token Estimator box to check that your token guess is realistic.
8. Switch to the Compare Models tab. Two rows are already there; click Add for a third and fourth. Set the shared usage numbers, then click Compare to get the table and bar chart.
9. Switch to the Pricing Browser tab, type at least one character (for example gpt, claude or llama), pick a sort field and direction, and optionally enable Reasoning models only. Click Use on any row to send that model back to the calculator.

Limits and Rules

• Pricing endpoints are rate limited to 120 requests per minute. Normal interactive use never comes close to this.
• Each token field accepts up to 50,000,000 tokens; larger values are silently clamped to that ceiling.
• Requests per month is clamped between 1 and 100,000,000.
• At least one token count must be above zero, or the server refuses to calculate.
• Comparison accepts at least 2 and at most 4 models per run.
• The pricing browser returns the top 200 matches, sorted by the field you choose.
• All prices are USD per 1 million tokens, taken from published list prices. Tiered volume discounts, batch-processing discounts and negotiated enterprise rates are not modelled.
• The pricing catalog refreshes every 24 hours. If a provider changed its rates today, the page may show yesterday rates until the cache expires, and the API marks that snapshot with a stale flag.
• No account, no API key and no captcha are needed. The calculator only reads pricing data; it never calls a model on your behalf.

When to Use It — and When Not To

Use it when you are picking a model for a new feature, checking whether prompt caching will pay for itself, forecasting next month spend, justifying a provider switch to your team, or explaining an invoice line to a client. It is also useful for spotting models whose output price is several times their input price, which matters a lot when your app generates long answers.

Do not rely on it when you need an exact invoice. The calculator uses published list rates only: batch discounts, long-context surcharges, cached-token promotions and per-seat subscriptions are not included. It also cannot price features the catalog does not publish rates for, in which case that line is simply omitted. For a final contract or a formal budget, confirm the numbers against the provider own pricing page.

Frequently Asked Questions