Prompt, history, files and tool definitions.
Includes thinking tokens.
Claude Opus 5.5
Same workload on every model
How the estimate works
Input tokens are split into the cached prefix and the rest. Cached tokens are charged at the cache-read price on hits and the cache-write price on misses; everything else at the base input price. Output tokens are charged at the output price, then the Batch discount is applied if selected. Claude Haiku 5.5 switches to its long-prompt price when a request is over 100,000 input tokens.
Not sure how many tokens your prompt is? Paste it into the token counter. Every model ID and price is also in the Claude models reference.
Frequently asked questions
How much does the Claude API cost?
You pay per million tokens (MTok), with separate input and output prices. For example Claude Opus 5.5 is $4 input and $20 output per MTok, and Claude Sonnet 5.5 is $2 and $10. The calculator above multiplies that by your real request sizes and volume.
How much does prompt caching save?
Cache reads cost a small fraction of the normal input price (0.1x on most models, less on the newest ones), while writing to the cache costs 1.25x for a 5-minute cache or 2x for a 1-hour cache. When the same long prefix is reused often, caching cuts input costs sharply; turn it on above to see your number.
What is the Batch API discount?
The Message Batches API charges 50% of the normal price for requests that can wait up to 24 hours. The discount stacks with prompt caching.
Do thinking tokens count as output?
Yes. Extended and adaptive thinking tokens are billed as output tokens, so include them in the output estimate for reasoning-heavy work.
Is Claude Code billed the same way?
With an API key, yes: Claude Code uses these token prices, and /cost shows the session spend. On a Pro, Max, Team or Enterprise plan, usage is covered by your subscription limits instead.
Sources: Claude API pricing. Claude Code Agents is an independent site, not affiliated with Anthropic.