Agent Cost Calculator
SourcesStart free

Anthropic source-linked planning

Claude API cost calculator

Model standard token usage and prompt-cache economics separately. The calculation keeps cache reads, 5-minute writes and 1-hour writes visible.

Provider calculator

Anthropic Claude API cost calculator

Model Claude text usage, cache reads, cache writes and retries with dated list rates.

Effective Verified Source ↗ Cache read, 5-minute write and 1-hour write are separate meters; the calculator exposes them separately.

Workload and token usage
Retry rate increases the number of billed attempts.
For models without a listed cache rate, these are billed at normal input price.
Extra billed attempts after failures or repairs.
Claude cache writes
Only used by records that list a 5-minute write meter.
Only used by records that list a 1-hour write meter.
Custom rate overrides
Enable this only when your contract or internal rate card differs from the selected public record.
Blank uses the record cache rate; if no cache rate is listed, normal input rate applies.
Assumptions

Rates are USD per million tokens unless a field says otherwise. This is a planning estimate, not a vendor quote. Records were verified 2026-11-02; region, contract, context-window, batch, tool and tax meters may differ.

Rate source ↗

No data is stored or uploaded.

Estimated monthly cost

$0.00

Enter assumptions to update the estimate.

Selected rate
$0.000Cost / request
$0.000Cost / task
0Billed attempts / month

Monthly token breakdown

New input$0.00
Cache read$0.00
5m cache write$0.00
1h cache write$0.00
Output$0.00

Estimate, not a provider quote. Public list rates, regional terms, contract discounts, taxes and usage meters can change. Check the linked source and replace assumptions with your account terms before budgeting.

Pricing and method sources

Rate cards display their own verification and effective dates. Use the original provider source to confirm current billing details.

Anthropic model pricing ↗Methodology and exclusions →

How this estimate works

Separate cache reads from cache creation

Claude API cost is not just uncached input plus output. A workload can create a prefix in the 5-minute or 1-hour cache and later read it. Those operations have different list rates. Enter measured or conservatively estimated tokens in each field; do not label all prompt tokens as cache hits unless your usage data confirms it.

Long prompts need the right model tier

The listed Haiku record is for prompts up to 100,000 tokens. Anthropic publishes a separate rate above that threshold. This page makes the short-context assumption visible beside the model and source. If the request crosses a threshold, use your provider’s matching rate or enter custom values.

For a broader unit-economics estimate, compare the result with the AI Agent Cost Calculator and track invoice differences with invoice reconciliation.

Price record and update boundary

This page uses 3 token-rate record(s) verified on 2 November 2026. The data are a snapshot, not a live quote. Every selected record includes its provider source link, effective date, verification date and context note.

Frequently asked questions

Does the Claude estimate include every API feature?

No. It estimates the displayed token and cache meters. Tools, search, audio, image input, batch, long context, taxes and negotiated terms may need separate treatment.

How should I estimate prompt cache hits?

Use observed cache-read tokens from provider usage where possible. Without telemetry, compare a no-cache case with a clearly labeled cache scenario.

What happens when my prompt is longer than 100,000 tokens?

The Haiku record in this tool is explicitly restricted to prompts up to 100,000 tokens. Select a matching public rate or enter your own above that threshold.

Is this a quote?

No. It is a planning calculation based on dated public rate cards and editable workload assumptions.

About the Author

Zunair Ali Syed is a doctoral researcher in Management Sciences at Minhaj University Lahore and a research consultant with experience in academic research and editing since 2011. He runs QuickEduHelp, with professional interests in digital learning, research methods, and evidence-based academic support.