Practical tools, and where to look inside them.

Break an AWS bill down by service, account and tag, and see how spend moved over time. The reference point for cloud cost habits you carry into AI spend.

The console reports are included with an AWS account. Programmatic API requests are charged per request.

Azure cost analysis, budgets and alerts, including Azure OpenAI usage.

Included at no extra charge for Azure usage. You still pay for the resources you analyse.

Cost reports, budgets and detailed billing export for Google Cloud and Vertex AI usage.

Reports are included. BigQuery billing export incurs normal BigQuery storage and query charges.

Where most real AI cost models start: rates in one column, volumes in another, cost per outcome at the end.

Free with a personal Google account.

The same modelling work with stronger pivot tables, common in finance teams.

Excel for the web is free with a Microsoft account. The desktop app needs a paid licence.

A fully open-source spreadsheet if you prefer not to use a cloud account for cost data.

Free and open source.

Try Gemini models, inspect token counts and compare prompt versions in a browser. The easiest starting point if you have no budget.

Has a free tier for experimentation. Paid usage applies beyond it.

Run prompts against OpenAI models and read the returned usage fields for input, output and cached tokens.

No permanent free tier. You need credits on the account, though small experiments cost cents.

Test Claude models, use the prompt tools and see token usage per request.

Usage based. Free trial credits are sometimes offered to new accounts.

Paste text and see exactly how it splits into tokens. The fastest way to prove a system prompt is longer than anyone thought.

Free, no account needed.

The open-source tokenizer library behind OpenAI models, for counting tokens in scripts and spreadsheets you build yourself.

Free and open source.

One interface across many providers, with routing, spend tracking and an openly maintained model pricing file that is useful on its own.

Open source and self-hostable. A paid managed offering also exists.

A single API across many models with published prices side by side, handy for quick price and capability comparisons.

Browsing prices is free, some models have free variants, and paid models are billed per use.

Trace every request, record tokens, latency and cost per workload, and compare prompt versions over time.

Open source and self-hostable, with a free cloud tier for small volumes.

A proxy and dashboard that logs requests, tokens and cost with very little code change.

Open source and self-hostable, with a free cloud tier.

Local tracing and evaluation for LLM and RAG applications, good for investigating retrieval behaviour.

Free and open source, runs on your own machine.

The vendor-neutral standard for emitting traces and metrics, including model calls, so your cost data is not locked into one tool.

Free and open source.

Tracing, datasets and evaluation runs, closely tied to the LangChain ecosystem.

Free developer tier with monthly trace limits. Team plans are paid.

AI request monitoring inside a wider enterprise observability platform, useful where Datadog is already the standard.

Commercial product with a time-limited free trial.

Vector search inside PostgreSQL, so retrieval costs sit in a database you already run.

Free and open source.

A purpose-built vector database with clear controls over index type and memory, which drive its cost.

Open source and self-hostable, with a free managed cluster tier.

A lightweight vector store that is easy to run locally while you learn how chunking affects retrieval cost.

Free and open source.

A managed vector database where storage and query volume map directly to a monthly bill.

Has a free starter tier. Larger indexes and throughput are paid.

Run the same evaluation set across models and prompts and compare quality and cost before changing production.

Open source command line tool, free to use. An enterprise offering exists.

Metrics for RAG answers, such as faithfulness and context relevance, so retrieval cuts can be judged on evidence.

Free and open source.

A test framework for LLM outputs, so a quality threshold can live in your pipeline rather than in an opinion.

Free and open source, with a paid hosted platform.

An AI cost management platform for allocating provider spend to workloads, tracking cost per outcome and reviewing optimization proposals with evidence.

Free tier plus paid plans for larger teams. Same idea as EC2 rightsizing, applied to AI models.

Open-source cost allocation for Kubernetes, useful when AI workloads run on your own clusters alongside provider APIs.

Free and open source, a CNCF project.

Policy as code for cloud resources, the usual way teams automate the Operate phase.

Free and open source.

Draw an AI architecture and annotate every component that produces cost, as in the Chapter 2 lab.

Free and open source, usable in the browser with no account.

Quick hand-drawn style diagrams for explaining a request path to finance or product.

Free and open source, with a paid collaboration product.

Shared whiteboarding when the architecture has to be annotated by several teams at once.

Free tier with a limited number of boards. Team plans are paid.

The open reference for FinOps capabilities, phases and personas that this course's language follows.

Free to read. Paid certification is separate and optional.

An open specification for a consistent billing data format across providers, which makes AI and cloud spend comparable.

Open specification, free to use.

Community guidance on applying FinOps practice to AI and machine learning spend specifically.

Free to read.