Why is my Anthropic Claude bill so high?
Anthropic gives you unusually good cost levers — a 90% cache-read discount, a 50% batch discount, and a wide Haiku-to-Opus price ladder. A high Claude bill almost always means one or more of those levers isn't being pulled.
Cache reads are 10% of input — are you using them?
Anthropic bills a cache hit at 0.1x the input price. If your workload re-sends a stable system prompt or document and your cache-read share is near zero, you're leaving the single biggest discount on the table.
Opus everywhere is expensive
Opus is priced for the hardest reasoning. Most production traffic runs fine on Sonnet, and simple tasks on Haiku at a fraction of the cost. The trick is routing only the easy slice down — guarded by an eval so quality never silently regresses.
Batch is 50% off and stacks with caching
For non-interactive work, the Batch API halves both input and output cost, and the discount stacks with prompt caching — compounding the saving.
Turn this into your number
Drop your usage export into the free analyzer and see how much of this applies to your account — provable savings separated from estimates. Nothing is uploaded.
Analyze my usage — freeFAQ
- Do you need my API key?
- No. You upload the Usage & Cost CSV from the Anthropic console and it's analyzed entirely in your browser. Nothing is sent to a server.