FREE · RUNS IN YOUR BROWSER · NOTHING UPLOADED

How much of your
Claude bill is waste?

Most production Claude workloads overpay 40–80% — prompt caching, the Batch API, model right-sizing and effort tuning are sitting unused. Describe your workload; get a lever-by-lever savings plan in dollars.

$ /mo
Requests sharing a stable prefix60%
Same system prompt / tools / context across requests → cacheable.
Traffic that could wait ≤1 hour25%
Nightly jobs, enrichment, evals, digests → Batch API.
Tasks a smaller model could handle35%
Classification, extraction, routing, short summaries.
ESTIMATED RECOVERABLE SPEND
$0
/ month · 0% of your bill · $0 / year
Levers are applied sequentially and conservatively (no double counting). Based on Anthropic list pricing: cache reads ≈ 10% of input price, Batch API −50%, Sonnet 5 intro $2/$10 vs Opus 4.8 $5/$25, Haiku 4.5 $1/$5. Effort tuning uses a conservative 15% output reduction on routine traffic — measure on your evals. This is an estimate, not an invoice.
Get this as an implementation plan

We'll send your numbers with the step-by-step checklist for each lever — cache breakpoints, batch migration, routing rules, effort settings.

The Claude Cost Playbook
Every lever above, engineered in depth — cache invalidators, batch patterns, routing evals, effort sweeps. $29 launch price.
→
5.maib.io — estimates from Anthropic list pricing · nothing you enter leaves your browser unless you submit the form
HubPlaybookmodels.json