AI Now Has a Volume Knob, and It Changes What You Pay

A couple of weeks ago we made the case that most AI announcements aren't worth your attention, and that's still true. A new model drops, the benchmarks tick up a point, everyone posts about it, and almost nothing changes for a normal business.

This one's different, and it's worth two minutes of your time.

On July 24, Anthropic released Claude Opus 5 with a feature that sounds boring and is actually a big shift: you can now tell the AI how hard to work. Low, medium, or high effort. A few days earlier, OpenAI shipped its GPT-5.6 family split into three tiers doing basically the same thing from the other direction. Two of the biggest labs, same week, quietly changing the same thing.

Here's what that means, minus the jargon.

What actually happened

Until recently, using a powerful AI model was mostly one setting: full blast. You asked it something simple, you got the same expensive, maximum-effort machine that you'd use for something genuinely hard. That's like starting your car and only having one speed, and it's a big reason AI bills have a habit of creeping up on businesses.

Anthropic's new toggle changes that. With Claude Opus 5 you can set the effort level per task:

• Low effort for quick, simple stuff (reformat this, draft a short reply, pull the key points out of an email).
• Medium for everyday work that needs a little thought.
• High for the genuinely tricky things where you want it to slow down, reason carefully, and check its own work.

Anthropic's pitch is that Opus 5 gets close to the quality of their top-end Fable 5 model at roughly half the price, while needing less back-and-forth. Take that "half the price" claim with the usual grain of salt (it's the company describing its own product), but the direction is real.

OpenAI got to the same place with a different design. Instead of one model with a dial, GPT-5.6 comes in three named versions: Sol (the flagship, most capable and most expensive), Terra (the middle), and Luna (the budget option). Same idea in a different wrapper. You pick how much horsepower you're paying for.

If you want the wider context on why the labs keep shipping like this, we covered the pattern in A New AI Model Drops Every Week. Most of It Isn't for You.

Why a normal person should care

Because this is the first model update in a while that touches your bill instead of a leaderboard.

Most AI announcements are labs competing on scores you'll never notice. This one is about control. For anyone who has watched an AI subscription or a per-use bill quietly climb, being able to say "don't spend Ferrari money on a bicycle errand" is the practical fix people have been asking for.

Think about how much of what you'd actually hand to AI is simple: cleaning up a paragraph, summarizing a thread, turning notes into a to-do list. None of that needs the most powerful, most expensive mode. Until now you were often paying for it anyway. The knob lets you match the cost to the job.

The other quiet win is speed. Lower effort usually means faster answers. For the small, high-volume tasks that make up most of a normal day, fast-and-cheap-and-good-enough beats slow-and-expensive-and-flawless almost every time.

What changes tomorrow

Honestly? For a lot of people, nothing you have to do by hand. If you use these tools through an app like ChatGPT or Claude, some of this happens behind the scenes, and the products increasingly pick a sensible effort level for you.

But three things are worth doing this week:

• Check whether your tools expose the setting. If you or your team use AI through the API or a business plan, there may now be an effort or model-tier choice you can set. Defaulting everything to "maximum" is how bills balloon.
• Stop treating "which AI is best" as the question. The better question is "which setting for which task." Best is now a range, not a single answer.
• Look at what you're actually spending. If AI costs have been fuzzy, this is a good moment to see where the money goes. A lot of it is usually simple tasks running on expensive settings.

This is the same principle behind not automating the wrong things: cheap, thoughtless scaling of the wrong setting just makes the waste bigger and faster. We got into that here: Automate the Wrong Thing and AI Just Speeds Up Your Mess.

What business owners should watch

A few things worth keeping an eye on over the next few months.

Your per-task cost, not just the sticker price. These tiers are priced very differently. OpenAI's budget option costs a small fraction of its flagship per unit of work. When you multiply that across thousands of little tasks, the setting you default to matters more than which brand you chose.

Vendors quietly defaulting you to the expensive mode. Some tools built on top of these models will pass the savings along. Others will keep you on the priciest setting because it costs them nothing and pads the invoice. If you're paying a company that runs AI on your behalf, it's fair to ask which model and effort level they use, and why.

"Good enough" getting genuinely good. The interesting part of Opus 5's claim isn't the top end, it's that the cheaper setting is now close enough for most work. If that holds up, the cost of doing the everyday stuff keeps falling, which is where most businesses actually feel it.

If you're evaluating a partner or platform right now, this is a great thing to press on. We put together the questions worth asking here: How to Pick an AI Automation Partner (10 Questions).

Hype, or does it matter?

It matters, but not in the way the headlines make it sound.

It's not a leap in intelligence. Nobody's daily life changes because a model scored higher on a coding test. What changed is the economics: powerful AI is getting cheaper to run for ordinary work, and you now have a lever to control the trade-off yourself. That's a slow, boring, real kind of important, the kind that shows up on a bill three months from now rather than in a demo today.

The honest caveat: dialing effort down too far on something that needed care will get you a fast, cheap, wrong answer. The skill this creates isn't technical. It's judgment about which tasks deserve the good setting and which don't. That judgment is the actual work now, and it's not something the toggle does for you.

If you're trying to figure out where AI actually fits into your business or daily workflow, and how to set it up so you're not overpaying for the simple stuff, that's exactly what we help people do at Humanity AI.