Only 22% of finance leaders can tie their AI spend to a business outcome, according to CloudZero’s 2026 finance survey. When AI ROI falls short, it usually isn’t because a company is doing too much AI. It’s that no one is watching which model runs which task, and that one choice accounts for a large part of the cost.
Here’s why it happens. Engineers and other AI users typically default to using the most capable model (hello, Fable) because it feels like the safe choice, and the safe choice is almost always the most expensive one.
That instinct may make sense in isolation. But the problem is scale. Multiply one over-powered model choice across every prompt, every agent, and every workflow, and you get a real number on the P&L that nobody chose on purpose. A default choice is not a decision. Being intentional should be the goal.
What CloudZero’s Model Rightsizer does
Today we open-sourced CloudZero Model Rightsizer and made it free for anyone to use.
It’s an agent you clone into your Claude Code workflows. It reads how your teams and agents actually use AI models, then sizes the model for your workloads up or down . It routes work to a smaller, cheaper model where it would deliver the same quality as the premium one you’re paying for. And conversely, it won’t let you starve a complex challenge with a low-tier model. It will right-size to a more capable model for those situations.
Internal teams already do this with cloud infrastructure. You don’t run a workload on an oversized instance and pay for headroom you never touch. You match the resource to the job. Model Rightsizer brings that same discipline to model selection, which is fast becoming one of the largest variable costs on an AI-forward company’s books.
Report
Finance needs to prove AI’s return: CloudZero report
260 senior finance leaders (more than half CFOs) told us why the speed of seeing AI spend, not the size of it, separates who pulls ahead on AI from who gets burned.
Lower the cost without lowering the output
Most cost levers ask the business to do less, like cut usage, cap tokens, or throttle a feature, and the number goes down because the work goes down with it. That’s a spending cut, and finance has seen enough of them to be suspicious.
Rightsizing is a different lever. It lowers cost without lowering output. You keep the same features, the same customer experience, the same throughput. You simply stop paying premium rates for routine work.
That distinction is the whole point. One path asks the business to do less. The other asks it to pay the right price for what it already does. And because every recommendation is specific and grounded in your own usage, finance can act on it without relitigating whether quality will suffer. That flows straight to gross margin and to the cost-to-serve number your board is starting to ask about.
How a company uses it
Picture a customer-facing summarization feature running on a top-tier model. Model Rightsizer shows that most of those calls are short, low-complexity requests a mid-tier model handles at the same quality for a fraction of the price. The team routes the simple calls down a tier and keeps the harder ones where they are. Cost-to-serve on that feature drops, margin improves, and the customer never notices a thing.
Repeat that across a dozen features and agents, and the savings compound into something worth reporting. Engineers and other practitioners run it where they already work, inside Claude Code, so cost context shows up next to the code instead of arriving weeks later with the bill. Finance sees the result where it matters, in cleaner AI unit economics.
Our own experience at CloudZero is telling: A colleague ran the agent and Opus spend dropped about 75%, from $2,813 to $711. At the same time, Sonnet usage jumped from 7 runs to 1,638, and those 1,638 runs cost $22 total. The same work ran on a cheaper model, hundreds of times over. (When we ran this to evaluate which of 11 tasks should have actually used Fable, it returned 0 of 11. Over-using Fable on those tasks would have been wasteful.)
Why we’re giving this away
CloudZero has spent years making one argument: AI without economics is just spending. We build the cost intelligence layer that tells a company not only what it spent on AI, but what that spend produced, mapped to the customers, features, and outcomes the business actually runs on. That’s the harder, more valuable half of AI ROI, and it’s the half we were built for.
Model Rightsizer is the first piece of that work we’ve put in anyone’s hands for free. We did it because we believe model right-sizing should be table stakes today. We saw that this could solve a basic problem that could help save companies money by simply automating model choices, so we decided to give away the easy lever to the community while we continue to focus on building out the harder, more valuable part in our platform – connecting every AI dollar to the outcome it produced.
Model selection is one of the clearest, most controllable levers in AI economics, and until now it has been nearly invisible. Making it visible, at no cost, is the fastest way to show the market what’s possible when you can actually see the economics: right-sizing is the first lever, and the board-level questions about unit economics and cost-to-serve are the ones right behind it.
Call it a first-mover’s move. We named the AI ROI problem early, we’re building the layer that answers it, and this is us handing over a working piece of it so you can see for yourself.
How to get it, free
Model Rightsizer is open source and free to download from CloudZero’s Skills Reference page. To get started:
- Open the Skills Reference page at docs.cloudzero.com/docs/ai-skills.
- Install the CloudZero Plugin for Claude Code from the linked marketplace. It’s free and open source.
- Connect your CloudZero organization context so the skill can read your actual model usage.
- Run the rightsizer skill, either with its slash command or by asking in plain language what your model spend looks like and where you’re overpaying.
- Review the right-sizing recommendations and act on the ones you trust.
The companies getting AI ROI right aren’t the ones spending the least. They’re the ones who can see the economics clearly enough to spend with intent. Model selection is one of the clearest levers you have, and it costs nothing to find out what you’d save.
Read more about how and why we built Model Rightsizer in this blog post by Djo Lopez.
