Home › Blog › The LLM Cost Audit Checklist

Checklist

The LLM Cost Audit Checklist

Use this checklist to audit your own LLM spend. Most teams find their first 20% in the first ten checks.

Quick answer

Audit in this order: where the money goes, caching, output, model choice, batch, agents and retries, then limits and measurement.

Where the money goes

  1. Spend by model.
  2. Spend by feature or call site.
  3. Input vs output share.
  4. Top customers by cost.

Caching

  1. Cached input share per model.
  2. Static content first in every prompt.
  3. No timestamps or IDs in system prompts.
  4. Tools in a stable order.

Output

  1. max_tokens set per call site.
  2. Answers no longer than needed.
  3. Reasoning effort matched to the task.

Models

  1. The cheapest model that passes evals, per call site.
  2. An evaluation set for each important task.
  3. No outdated, more expensive models left running.

Batch

  1. Background jobs on batch APIs.
  2. Evaluations on batch.

Agents and retries

  1. Max steps per task.
  2. Per-task budgets.
  3. One retry layer with backoff.
  4. Timeouts above p99 latency.
  5. Duplicate-request detection.

Limits and measurement

  1. Budgets and alerts.
  2. Per-customer caps.
  3. Cost per successful task tracked weekly.
  4. Dashboard reconciled with invoices.

Want it done for you? The free 15-minute teardown runs this against your usage export.

Frequently asked questions

How do I audit my LLM costs?

Break spend down by model, feature and token type, then check caching, output length, model choice, batch eligibility, agent limits and retries.

How long does an LLM cost audit take?

A first pass takes an hour with a usage export. Our free teardown takes 15 minutes.

Want us to find your 20%?

Book a free 15-minute API cost teardown. Bring last month's provider usage and we'll show you your biggest waste and the first fix we'd ship.

Get a Free 15-Minute API Cost Teardown

Related

Free LLM API Cost AuditA free 15-minute LLM API cost audit for OpenAI, Anthropic and Gemini users: we size your avoidable spend from a usage export and show the first fix we'd ship..Why Is My OpenAI API Bill So High? A 60-Minute DiagnosisWhy your OpenAI API bill is so high and how to find the cause in an hour: output tokens, uncached prompts, retries, agent loops, the wrong model and missing batch..Bill X-ray: LLM Usage CSV AnalyzerUpload or paste your LLM usage export and see spend by model, cache share and output share, with the red flags that point to savings.