LLM Cost Calculator
Estimate your monthly LLM API bill and compare models. Set your usage and see the cost across providers.
Estimated monthly cost at 60,000 requests/month:
Approximate public list prices (USD/1M tokens); actuals vary by version, batching, caching & discounts. Prompt caching and model routing can cut these 40–80% — something we design into every build.
How the LLM cost calculator works
A tool that will not show its arithmetic is worth less than one that does.
- 1
You enter expected requests per month, and the average input and output tokens per request.
- 2
Those are multiplied by the selected model’s published per-million input and output prices — which differ, usually with output costing several times more.
- 3
The result is a monthly run rate, shown per model so the same workload can be compared across providers.
What it assumes
Published list prices. Committed-use discounts, batch pricing and enterprise agreements are not modelled.
Roughly four characters per token for English. Code, JSON and non-Latin scripts run denser, sometimes considerably.
One call per request. Anything agentic — retries, tool loops, self-correction — multiplies real usage several times over.
No prompt caching. Where a long system prompt repeats across calls, caching can cut input cost substantially and is not reflected here.
How to read the result
The number that usually surprises people is how much of the bill is output tokens. If your estimate looks unaffordable, the first lever is almost never a cheaper model — it is asking for shorter output, then caching the repeated part of the prompt, then routing easy requests to a smaller model and hard ones to a large one. Reach for a cheaper model last, because that is the change most likely to cost you quality.
Questions
- Why is my real bill higher than this estimate?
- Almost always retries and multi-step calls. One user-visible action in an agentic system can be five or ten model calls. Count actual calls, not user actions.
- How do I cut cost without changing model?
- Cap output length, cache the static part of your prompt, trim retrieved context to what is actually used, and route by difficulty. In practice those four together tend to matter more than the choice of provider.
- Are these prices current?
- They are the published list prices at the time of writing and providers change them, usually downward. Check the provider’s pricing page before committing to a budget.