Anthropic invoices hide insights that can directly impact your Claude spending.
When reviewing your Claude usage, focus on four key line items: input_tokens, output_tokens, cache_read_input, and cache_creation_input. Input_tokens track the data you send to Claude, output_tokens measure the responses generated, while cache_read_input reflects hits from prior requests, and cache_creation_input covers building that cache. These metrics reveal where costs accumulate, especially in repetitive or iterative workflows.
For example, in case-001, an indie hacker achieved identical results on CRM UI prompts, dropping their bill from $312 to $74 by optimizing based on these lines. Likewise, case-006 demonstrated a developer reducing expenses from $145 to $29 for daily coding assistance, maintaining quality. And in case-002, a small SaaS team saw savings from $1840 to $287 on customer support replies. To apply this, visit aiusage.ai and examine your own patterns.
By auditing these line items, you can identify inefficiencies without changing your prompts. Try it—paste your last 30 days at aiusage.ai for a clear breakdown.
Audit your own Claude usage. Paste your last 30 days at aiusage.ai — no signup for the number.