The price of AI just fell off a cliff. That is the best news your business got this summer.

It has been about a month since I posted here. That was not the plan. Client work won, the way it does when you run the shop and also do the work. But the gap turned out to be useful, because the thing that happened while I was heads down is the thing this blog exists to cover. The price of running AI fell, hard.

On July 30, OpenAI announced major price cuts across its models, with one of its main models dropping roughly 80 percent and another around 20 percent. That is not a coupon. That is the raw material most AI products are built on getting cut by four fifths in a single announcement.

Here is my position, and it is why this blog has a point of view instead of a news feed. The cost of intelligence is collapsing, competition is what is doing it, and for a normal business that is the best news of the year, as long as you stop paying last year's prices out of habit.

Let me steelman the worry first, because there is a serious one. Forbes argued these cuts could trigger a race to the bottom. The concern is fair. If frontier labs are selling access below what it costs them to serve, burning investor money to buy market share, then today's price is a loss leader and tomorrow's price is whatever they want once you are locked in. Cheap now can mean captured later. Anyone who watched rideshares go from subsidized to expensive knows the pattern.

That worry is real, and it is exactly why the important story is not OpenAI's discount. It is why they had to offer it.

They had to because of open-weight models. Forbes' own follow-up called this the efficiency era and told executives to learn open weights for a reason. Models you can download and run yourself, from DeepSeek, and the GLM, Qwen, and MiniMax families, now do real work at a fraction of frontier prices. When a capable model is free to download and cheap to self-host, no proprietary vendor can charge frontier rents for routine work.

That is also the answer to the race-to-the-bottom fear. A price cut a single vendor can reverse is a promotion. A price floor set by a free model your competitor can also run is structural. The second one does not un-happen when the marketing meeting ends. So yes, be skeptical of any one lab's discount. Do not be skeptical of the open-weight competition forcing it, because that is the part you can actually rely on.

I will say the free-market piece plainly, because it is a genuine conviction and it is relevant here. No mandate, no committee, and no regulation delivered this price drop. Rivalry did. The most useful thing a policymaker could do for AI affordability is stay out of the way of the open-weight ecosystem already driving prices down. Heavy compliance rules do not hurt the trillion-dollar labs that can afford lawyers. They hurt the small teams shipping the cheap alternatives, which means they hurt you, the business trying to buy AI without overpaying.

So what do you actually do with this, if you run a business and not a lab.

Stop paying for a Ferrari to drive to the mailbox. Most of what a business automates is routine. Sorting inbound email, drafting first passes, pulling data out of documents, answering the same twenty questions. That work runs fine on a small or open model for a small fraction of frontier pricing. Reserve the expensive model for the genuinely hard minority of requests. That split, a cheap model for the routine ninety percent and a strong model for the hard ten, is the highest-impact cost decision most companies are not making. They picked one expensive model in 2025 and never revisited it while the ground moved under them.

That is the beat this blog is back to cover. Where the money leaks, which model is actually right-sized for the job, and how to spend less without shipping worse. It is also, not by accident, the work I do. MojoAI exists to right-size this for businesses that do not have time to track four model families and a price cut a week. If your AI bill has climbed while the market's prices fell, that gap is money, and it is worth an afternoon to close.

If you want a straight read on what you are overpaying for, tell me what you are running. Good to be back.