LLM cost calculator
Estimate the API cost of a request and a month of traffic across 30 models. Runs in your browser — nothing you type is uploaded.
Token counts are approximated at ~4 characters per token. Use your provider's tokenizer for exact billing.
| Model | Cost / request | Cost / month | Tokens / month |
|---|
Monthly spend by model
How to read the numbers
Cost per request is input tokens × input price + output tokens × output price, at published list prices. Cost per month multiplies that by your daily request count and by 30.
Output tokens are usually the expensive half. On most models output costs four to eight times input per token, so the shape of your workload matters: a short-prompt, long-answer workload behaves very differently from a long-context, short-answer one.
The three numbers that decide your bill
- Input tokens per request — dominated by conversation history and retrieved context, not by what the user typed.
- Output tokens per request — cap it. A max-token setting that is far above typical output is a silent overspend.
- Requests per day — includes retries, evals, tool loops and background jobs, not just user-visible calls.
Want this in your own dashboard?
The Pro pricing API returns the same numbers as JSON, so you can price your own workloads automatically.
Pro pricing API docsFrequently asked questions
What counts as an input token?
Everything you send: the system prompt, retrieved documents, tool definitions, and the whole conversation history on every turn. In multi-turn chat it usually dwarfs the visible message, which is why the input box defaults to 1,500 tokens rather than 100.
Why is my real bill higher than this estimate?
Three common reasons: you send more input tokens than you think (history and retrieved context), you rarely hit cache, and retries or tool-calling loops multiply the request count. Also check whether you are billed at a long-context tier, which some vendors price higher above a threshold.
Does caching change the estimate?
Yes, and it can be dramatic — cached input is often around 10% of the standard input rate. Model pages list the cached input price where the vendor publishes one. Subtract your expected cache-hit rate from the input side to model it.
Is there a difference between batch and real-time pricing?
Often yes; several vendors discount asynchronous batch jobs. We show standard list prices. If you can run offline, check the vendor's batch tier before ruling a model out on price.
How current are the prices?
The dataset was last refreshed Oct 15, 2026, and every model page links to its source. We re-verify when vendors announce changes; the changelog records what moved.