Am I overpaying for LLM APIs? A 2-minute self-check
Most teams are overpaying by a meaningful margin, and the causes are consistent. Run through this checklist; if you tick two or more, there's recoverable money in your bill.
The checklist
Is a large system prompt or document sent on most calls? (caching)
Does most spend run on your provider's top model? (right-sizing)
Do you run nightly or bulk jobs synchronously? (batch)
Is max_tokens unset on most call sites? (output bloat)
Are you still calling a deprecated model snapshot? (stale model)
Turn the checklist into a number
The free analyzer reads your usage export and attaches a monthly euro figure to each of these — the defensible floor separately from the estimated ceiling — so you know what's actually worth fixing.
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 — free