concept: token-based pricing

Token-based pricing

The Anthropic API bills usage based on the number of tokens processed, not a flat per-request fee -- both input and output tokens count toward billed usage.

Input vs output rates

Most LLM API providers, including Anthropic, bill output tokens at a materially higher per-token rate than input tokens, because generating tokens is more computationally expensive.

Prompt caching lowers effective cost

When a prompt prefix is cached (see subagentcontext.com's prompt-caching concept), reusing that cached prefix is billed more cheaply than reprocessing fresh each time.

No specific dollar-per-token figures are stated on this page.