Free AI Model Price Comparison
Compare by cost per task, not by price list
Price lists per million tokens are hard to compare because input and output are priced differently, and because the cost of a task depends on how many tokens it uses. A model that looks cheap per token can be expensive if your task is mostly output. This tool takes your actual usage, input tokens and output tokens per request and requests per month, and calculates the monthly bill on each listed model from Anthropic, OpenAI and Google, so the comparison is about your work and not a table of rates. Everything runs in your browser.
How to use it
- Enter input tokens per request, including your system prompt and any context. The AI token counter helps you measure them.
- Enter output tokens per request, the typical length of the reply.
- Enter requests per month.
- Filter by provider if you only care about one company.
- Edit any price to match your contract or a newly announced rate. Reset prices restores the reference values.
- Copy the table as text for a proposal or a budget.
What shapes your result
- Input-heavy work, such as summarising long documents or answering from big passages, depends mostly on the input price.
- Output-heavy work, such as writing articles or generating code, depends mostly on the output price.
- High volume of tiny requests, such as classifying short messages, rewards the cheapest capable model.
- Rare, difficult tasks justify a larger model, since quality matters more than the small bill.
Small models and large models
Every provider sells a ladder: small, fast, cheap models at the bottom and larger, more capable, costlier ones at the top. The sensible approach is to use the smallest model that does the job well. Test a sample of your real inputs on a small one first, and move up only when the results are not good enough. Many teams also route requests, sending the easy majority to a small model and escalating only the hard ones, which can cut costs dramatically with little loss in quality.
What the comparison leaves out
- Discounts for prompt caching, batch processing or committed spend.
- Extra charges for long-context tiers, images, audio, web search or tool use.
- Taxes and currency conversion.
- Quality differences. The cheapest model is only the cheapest if it does the job.
If you use caching, lower the input price you enter to reflect the cached share.
Keeping the figures honest
The reference prices are maintained by hand, and the date they were last checked is shown in the tool. Providers change prices and release models often, so confirm current rates on each official pricing page before you commit a budget. When a price changes, edit it in the table and the whole comparison updates. A quick review every month or two keeps your estimates trustworthy.
Reading the table
Each row is a model, with its provider, its input and output price per million tokens, and two results computed from your numbers: the cost of a single request and the cost of a month. Use the Show menu to look at all providers, or only Anthropic, OpenAI or Google. The prices in the table can be edited, so if you have negotiated a different rate, or a provider has announced a change, type the new figure and the comparison updates. Reset prices returns everything to the reference values, and Copy as text gives you the table to paste into a document.
A worked example with invented prices
Take a feature with 2,000 input tokens and 500 output tokens per request, run 100,000 times a month. At a made-up price of $1 per million for input and $4 for output, one request costs $0.004 and the month costs $400. At $3 and $15, the same usage costs $1,350. At $0.25 and $1, it costs $100. Whether the cheaper model is good enough is a quality question that only testing can answer, but now you know exactly what is at stake.
Related tools
For a single model in detail, with per-day and per-month figures, use the AI API cost calculator. To check whether your text fits a model’s limits, try the AI context window checker.
FAQ
It depends on your usage. Small models from each provider are usually cheapest. Enter your token numbers to compare for your case.
They are reference prices checked on the date shown in the tool. Edit them if a provider has changed its rates.
Generating text takes more computing work than reading it.
No. Adjust the input price yourself to reflect cached input if you use it.
No. The calculation runs in your browser.
Yes. Edit any input or output price to match your own contract or a new announcement, and the costs update. Reset prices restores the reference values.