LLM API cost calculator

How to use it

1. Pick a model. Its official input, output and cached-input prices fill in automatically, and you can edit them.
2. Enter the average input and output tokens per request and how many requests you expect per day. Not sure about tokens? Paste a typical prompt and the calculator estimates them.
3. If you reuse a long system prompt or shared context, set the share of input that is cached. Providers bill cached input at a fraction of the normal price.
4. Read the cost per request, day, month and year, then compare the same workload on every model in the table below.

What actually drives LLM cost

• Output tokens. Output is priced about five times higher than input on most models. Asking for shorter answers, or structured output instead of prose, is usually the biggest single saving.
• Context size. Retrieval systems often send thousands of tokens of retrieved text with every question. Tighter retrieval and smaller chunks cut input cost directly.
• Caching. A long system prompt sent with every request is the textbook case for prompt caching. At a 10x discount on cached tokens, a mostly cached prompt costs a fraction of the list price.
• Model routing. Most production systems route easy requests to a small model and only send hard ones to a flagship. The comparison table shows how wide that price gap is.
• Long-prompt surcharges. Some models charge a higher rate once a single prompt passes a threshold (for example 100K, 200K or 272K tokens). The note under each model lists any surcharge.

How the token estimate works

Tokenizers differ by provider. For English text, OpenAI documents about four characters per token, roughly three quarters of a word. Anthropic documents that current Claude models produce more tokens for the same text, about 2.5 characters per token. The estimator uses the ratio for the provider you pick. For exact counts, use the provider's token-counting endpoint on your real prompts.

From estimate to production

API spend is rarely the largest line in an AI budget. The engineering work around the model usually costs more: retrieval, evaluation, guardrails, monitoring and the integrations that put an AI system into a real workflow. Our AI development cost calculator prices that side, the senior team that builds and runs it.

FAQ

How much does it cost to run an LLM in production?

It depends on tokens and volume. A support assistant handling 10,000 requests a day with 2,000 input and 500 output tokens each costs roughly $135 a month on the cheapest small models and more than $10,000 a month on the most expensive flagships, at list prices without caching. Enter your own numbers above to see the range.

Are these prices current?

They are official list prices for the standard tier, checked on each provider's pricing page on the date shown under the calculator. Providers change prices often, and some have announced changes, so check the linked source before you commit a budget.

Why is output more expensive than input?

Generating each output token takes a full pass through the model, while input tokens are processed in parallel. That is why output is priced several times higher and why shorter responses save the most.

What is prompt caching?

When the start of a prompt repeats across requests, such as a long system prompt or a shared document, providers can store it and bill the repeated part at a discounted cached-input rate.

Is my prompt sent anywhere?

No. The calculator and token estimate run entirely in your browser.