How to estimate your Claude API bill
Start with one typical request. Count its input tokens (system prompt, conversation history, documents and tool definitions) and the output tokens you expect back. Enter both numbers, then your monthly request volume. The table ranks every model by monthly cost so you can see the price difference between, for example, a Haiku and an Opus workload at a glance.
Not sure about your token counts? Paste a sample into the LLM token counter for a quick estimate, or call the count_tokens endpoint for an exact number.
Ways to lower your cost
- Pick the cheapest model that meets your quality bar. Test on real examples before you commit; the price spread between model tiers is large.
- Use prompt caching when a long, stable prefix such as a system prompt or document is sent on many requests.
- Use the Batch API for anything that can wait, for a 50 percent discount.
- Control output length. Output tokens cost several times more than input tokens, so concise answers and sensible
max_tokenslimits save money. - Tune effort. Lower effort settings use fewer thinking tokens on simple tasks.
What the calculator leaves out
It uses list prices and ignores prompt caching, tool-use overhead, server tools such as web search, fast-mode premiums, taxes, and volume agreements. Treat the monthly figure as a planning estimate and check your actual usage in the Anthropic Console.