Interactive tool
LLM cost calculator
Enter the token volume you expect each month and see what every model would actually bill you. Caching and batch discounts are applied where providers offer them.
Pricing checked against provider documentation on . How we verify
Share of input tokens served from cache. Chat and agent apps often reach 70%+.
—
| Model | Input cost | Output cost | Monthly total | vs cheapest |
|---|
Estimates use standard published rates and exclude tool-call fees, long-context surcharges, and committed-use discounts. Reasoning models will bill above these figures because thinking tokens count as output.
Frequently asked
How do I estimate tokens from my actual usage?
As a rough guide, one token is about four characters of English, so 1,000 tokens is roughly 750 words. Multiply your average prompt and response length by your monthly request count. For agent workloads, remember each task makes many model calls.
Why is my real bill higher than this estimate?
Three usual causes: reasoning models spend hidden thinking tokens billed as output, agent loops make far more calls than expected, and long prompts resend the same context every turn. Prompt caching addresses the third.
What is prompt caching worth?
Cached input typically costs about a tenth of the base input rate. For chat and agent workloads that resend a large system prompt each turn, this is usually the single largest saving available — often larger than switching models.