TokelyTHE SHORT ANSWER
Claude API cost depends on the selected model and separate charges for input, output and caching. As checked on October 15, 2026, Anthropic lists Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens, and Opus 5.5 at $4 and $20. Service tiers and context conditions can change the applicable bill. Official Claude pricing.
How much does a normal Claude request cost?
For an uncached Sonnet 5.5 request containing 10,000 input and 2,000 output tokens, the listed base rates produce $0.02 of input plus $0.02 of output: $0.04 total.
At 10,000 identical requests, that becomes $400 before additional charges or discounts. This is a tariff calculation, not a measured average request or a forecast for all Claude applications.
Output length matters. If those requests produce 6,000 output tokens instead, the per-request cost becomes $0.08 at the same input length.
How are Claude cache writes and reads priced?
The checked Sonnet 5.5 tariff lists a five-minute cache-write rate of $2.50 per million tokens and cache reads at $0.10. A one-hour write has a different rate. Use the current model-specific table rather than assuming every Claude model uses the same read multiplier. Anthropic pricing.
For a repeated 20,000-token prefix, one five-minute write costs $0.05. Nine reads cost $0.018. Ten uses of the prefix therefore cost $0.068, compared with $0.40 if processed at the regular input rate every time. The calculation assumes all nine reads qualify before expiry. Output and changing input remain separate charges.
What does a caching estimate need to include?
Measure cacheable prefix length, eligible reuse count and expiry. A theoretical hit rate is not a billing result. Read the reported cache usage and reconcile it with the invoice.
The prompt-caching guide explains eligibility and implementation. It is especially relevant for long system prompts, shared documents and tool definitions.
How does Tokely pricing compare?
Tokely lists Sonnet 5.5 at $1.50 input, $7.50 output and $0.075 cache reads per million tokens. The earlier uncached 10,000-input/2,000-output request would cost $0.03 at those published rates, 25% below the $0.04 direct base-rate calculation. Tokely Sonnet 5.5.
This comparison excludes cache writes, special tiers, taxes and any route-dependent price adjustment. Compare complete token categories for a real cached workload rather than extending an uncached percentage automatically.
Does a Claude subscription include these API calls?
Do not treat a consumer subscription as interchangeable with pay-as-you-go API access. Check the billing arrangement attached to the credential your application uses, particularly when connecting Claude Code through a gateway. Claude Code gateway billing.
Estimate your own request volume with Tokely’s calculator, then use the Messages API guide for supported request formats and cache fields.