AiGanakFind out what your LLM bill is really paying for.
We measure your OpenAI, Anthropic, Azure, Bedrock or Gemini usage from your own logs, test cheaper setups on your own approved answers, and hand you a report with every saving priced and every risk ranked.
The LLM Cost & Reliability Audit measures your OpenAI, Anthropic, Azure, Bedrock or Gemini usage from your own logs, prices every saving (model routing, caching, prompt trimming, retrieval, batch) against your approved answers, and ranks reliability risks. A free Quick Check gives an instant estimate.
Estimate your savings
Answer a few questions and paste a system prompt. Everything runs in your browser. Nothing is sent until you ask for the full audit.
An estimate from your answers and typical results. The audit measures it from your own logs.
What you get: a report like this one.
A fictional support-software company spending $14,001 a month. The audit found $5,851–$7,377 a month in savings (42–53%) through 9 changes, and 4 reliability issues to fix.
- Where the money goes, by feature and by part of each request
- Every saving priced, with a conservative and a full case
- Model tests on your own approved answers
- Prompt-injection and format failures ranked by severity
- A 30/60/90-day plan and a self-hosting break-even check
Simulated data. Your results depend on your workload.
From export to report in five to seven working days.
Send your data
One month of usage export, a sample of logged requests with personal data removed, and approved example answers. Under NDA.
We measure
Our analysis engine prices every change from your data. Model Lab replays your examples on cheaper models. Prompt Lab tests your prompts.
Architect review
An AiGanak architect checks every figure and recommendation before the report is released.
Walkthrough
You get the report and a 45-minute call. We can then implement the plan or hand it to your team.
Choose the depth you need.
Instant estimate and prompt check on this page.
Usage export and up to 500 logged requests, up to 2 features. Full report and walkthrough.
Full logs, model replay on your examples, up to 5 features, prompt-injection testing.
We implement the plan, then re-audit every quarter from $500 a month.
Nine levers, priced in order so nothing is counted twice.
| Lever | What it means | Typical place it shows up |
|---|---|---|
| Route to smaller models | Send simple requests to a cheaper model when it scores within one point on your examples | Classification, simple Q&A, summaries |
| Retrieve fewer documents | Cut retrieved context to what answers actually use | RAG assistants and help centres |
| Trim system prompts | Remove repeated instructions and surplus examples | Long, many-example prompts |
| Provider prompt caching | Make prompt prefixes stable so they bill at the cached rate | High-volume real-time features |
| Cache repeat answers | Serve identical questions from a cache | FAQs and help widgets |
| Batch pricing | Move work that can wait to discounted batch processing | Nightly summaries, reports, backfills |
| Shorter answers | Set length and format to what users need | Drafting and summarising |
| Agent memory and step caps | Summarise history and stop runaway loops | Tool-using agents |
| Structured outputs | Use JSON schema mode to stop retries | Extraction and classification |
Data, security and results
Can't find it? Press ⌘K and ask.
What data do you need for an LLM cost audit?
One calendar month of usage export from your provider's console or usage API, a sample of 500 to 5,000 logged requests and responses with personal data removed, and 20 to 50 approved example answers per feature. You run the export yourself, so we never need your API keys.
Which providers are supported?
OpenAI, Anthropic, Azure OpenAI, AWS Bedrock, Google Gemini and self-hosted open models. The importer reads the OpenAI and Anthropic usage APIs directly and a simple CSV for the rest.
How do you make sure quality does not drop?
Every model change is tested on your own approved answers in Model Lab. A cheaper setup is only recommended if it scores within one point of what you use today on a 0 to 100 scale.
How much does an LLM cost audit cost?
The Quick Check is free. The Standard Audit is $1,500 for up to 2 features. The Deep Audit is $4,500 for up to 5 features with model replay and prompt-injection testing. Implementation starts from $5,000.
How long does an audit take?
Five to seven working days from receiving your data, including an architect review and a 45-minute walkthrough call.
Is my data safe?
Work is done under NDA, on redacted samples only, in an isolated environment, and the data is deleted 30 days after delivery.
Get the full audit for your setup.
We reply within one working day with a data checklist, an NDA and a fixed quote.
Run the Quick Check to attach your estimate.