Guides to cutting your LLM API bill
15 guides on where the money goes in an AI project and how to spend less of it. Every figure is checked against the providers' published rates. If you are starting from nothing, read the first three in order.
Got a shorter question? The FAQ answers eighteen of them, from what a chatbot costs to what to charge a client.
Start here
Three guides, in order, that cover most of what a first project needs.
- Understanding LLM Token Economics: Inputs, Outputs, and Context Windows
Input and output are metered separately, and output usually costs several times more. Why the shape of a workload decides the bill, not its size.
2 min read · Start
- How Much Does It Cost to Run an AI Chatbot?
The same chatbot can cost $5.04 or $1,080 a month depending on the model behind it. A worked example, and the four levers that close the gap.
2 min read · Start
- Cheapest LLM API: How to Actually Compare Prices
There is no cheapest model, only a cheapest one for your shape. Three workloads, three different winners, priced on rates we recorded ourselves.
4 min read · Start
Quoting and selling this work
For consultants and agencies putting a number in front of a client.
- How to Price an AI Automation Project
An AI project has three costs and only one behaves like software work. How to size the running cost before you build, and what to put in the quote.
2 min read · Start
- How to Explain AI Costs to a Non-Technical Client
Your client is not asking what a token is. They are asking whether this will surprise them. Metered like electricity, not licensed like software.
2 min read · Start
- Quoting Fixed-Price AI Work Without Eating the Variance
Fixed price and a metered dependency do not mix. Three structures that hold, and how to set a cap from the number you do not expect to pay.
2 min read · Practical
Cutting a bill you already have
The levers, roughly in order of how much they move.
- How Prompt Caching Cuts LLM API Bills by Up to 80%
Providers charge a fraction of the input rate for a prefix they have already seen. What prompt caching saves, and how to structure a prompt to earn it.
2 min read · Practical
- Prompt Optimization: The Cheapest LLM Cost Cut You Are Not Making
Before you switch models, edit the prompt. Filler, duplicated instructions and stale context are billed on every single request, forever.
2 min read · Practical
- Batch APIs: How Much You Save by Agreeing to Wait
Waiting is the largest discount available on an AI bill. Which of your calls could have waited, how to size the saving, and what to verify first.
4 min read · Practical
- Anthropic Prompt Caching: A Step-by-Step Audit of Your API Bill
Claude reads cached input at a median of 90 percent below its standard rate. A four-step audit of whether your prompts actually earn it.
5 min read · Detailed
Auditing what you actually paid
Start from your own usage export rather than an estimate.
- How to Audit Your OpenAI or Anthropic Bill (and Cut It)
Export your usage, price what you actually spent, then re-price the same tokens on every other model. How to audit an OpenAI or Anthropic bill.
2 min read · Practical
- How to Export and Read Your OpenAI Usage CSV (and Find the Hidden Costs)
Where to find OpenAI usage export, how to read each column, and the three places hidden costs collect in it once you start looking.
2 min read · Practical
Choosing and measuring
Comparing providers, pricing a pipeline, counting what you send.
- OpenAI vs. Anthropic vs. DeepSeek: A Developer Pricing Breakdown
OpenAI, Anthropic and DeepSeek price for three different strategies. How the tiers compare, and why production teams end up routing across all three.
2 min read · Start
- FinOps for AI: Calculating the Real Cost of Your RAG Pipeline
A RAG query looks small and is not: retrieved context is a median of 86 percent of the bill. Retrieve less, cache the rest, price the layers.
4 min read · Detailed
- Free AI Token Counter: Paste, Count, and Price Your Prompt
Paste a prompt, document or code file and watch the token count update as you type. Nothing is uploaded. Then price the same text on every model.
2 min read · Start