The price of using AI moved under a lot of businesses this month, and most owners did not notice. OpenAI put GPT-5.6 out to the public on July 9 with three differently priced tiers and a new agent that runs whole tasks and charges as it works. Newer models count tokens differently than last year’s did, so the exact same document can quietly cost more to process. None of these shifts is dramatic on its own. Stacked together, they turn the thing you budgeted as a flat monthly tool into a variable one — and a variable cost nobody is reading is how surprise bills happen.

The flat fee is going variable

A per-seat subscription is easy to budget: fixed number, same every month. But more of what you actually use now bills by consumption — per token, per task, per agent run. When you point an AI agent at a job and let it work for an hour across your files, you are metering. When your team triples its usage because the tool got genuinely useful, the number climbs even though nothing about the pricing changed. That is not a trap; it is just a different shape of cost than the one most small businesses are set up to track.

What actually drives the number

Three things move your AI bill: how many people use it, how heavily each one uses it, and which tier they reach for. A cheap fast model and a top-tier reasoning model can differ by five to ten times on the same task. Most work does not need the expensive tier. If your team defaults to the biggest model for everything — including drafting a two-line email — you are paying flagship prices for grunt work. Knowing your split between light and heavy usage is most of the battle.

Set up spend visibility in an afternoon

You do not need a finance system for this. Most AI platforms now expose an admin dashboard with usage and spend by user, plus alerts you can set at a dollar threshold. Turn those on. Then put one recurring line in your monthly close: what did AI cost us, and what did we get for it. The second half matters as much as the first — a rising bill tied to real output is an investment; a rising bill nobody can explain is a leak.

The honest caveat

Watching spend is not the same as cutting it. The goal is not the smallest possible bill — it is knowing the number and deciding on purpose. Sometimes the right move is to spend more, on a tier that does the job in one pass instead of three cheap tries. Visibility just means you are choosing that, not discovering it on an invoice.

The takeaway

Before you scale AI across your business, make its cost a number you look at every month, next to revenue and payroll. Set spend alerts today, name a person who owns the review, and decide your tiers deliberately. The businesses that get burned by metered AI are the ones who treated it like a flat subscription right up until the month it wasn’t.