ANTHROPIC CLAUDE API · USAGE-AWARE
Claude API Cost Calculator
Estimate Claude Sonnet 5.5 costs from writing, summarization, coding, or research tasks—or enter measured tokens for caching, Batch, and inference-geography pricing.
Tell us what you do
Based on observed usage samples and official pricing. Switch to measured usage when you have response usage fields.
Compare estimated cost ranges
These ranges use each measured model’s observed usage for the same task, size, and count. They do not compare quality or speed.
CLEAR PRICING MATH
How Claude API cost is calculated
Each usage category stays separate, so cache writes and cache hits are not mistaken for ordinary input.
input + output + cache writes + cache readsBatch applies a 50% discount. US-only inference applies a 1.1× multiplier to all token categories.
Input and output
Input and generated output use separate per-million-token rates. Copy actual counts from the API response for a workload-specific estimate.
Prompt caching
A 5-minute write costs 1.25× base input and a 1-hour write costs 2×. Cache reads are cheaper, but savings require enough reuse.
Batch processing
Anthropic publishes a 50% discount for asynchronous Batch workloads. It is a cost option, not an equivalent low-latency replacement.
QUICK EXAMPLES
Example monthly workloads
1M input + 200k output
$4.00 / month1M input + 200k output
$2.00 / month100k 5m write + 1M cache reads
$0.02 / monthFor the most accurate result: use input_tokens, output_tokens, cache_creation_input_tokens, and cache_read_input_tokens from the Claude API response. Free credits, taxes, and negotiated pricing are not treated as public fixed rates.