Claude Sonnet 5 Is Cheaper. That's Not the Real News.

On Monday, Anthropic released Claude Sonnet 5. Most of the coverage led with the price, and fair enough. It runs close to their top model at a fraction of the cost, and it is now the default for anyone on a free or paid Claude plan.

Cheaper and smarter at the same time. That is a good week for the buyer.

But the price is not the part I want you to notice.

Here is the part. Sonnet 5 is the default. You did not choose it. If you use Claude, or a tool built on Claude, the model answering you this week is better than the one that answered you in the spring. Nobody sent you a memo. The floor just moved under your feet.

This happens more than people track. Every few months, the AI inside the tools you already pay for gets quietly better. Not because you upgraded. Because the vendor swapped the engine and kept the same dashboard.

That changes how you should treat a decision you probably made months ago and forgot you made.

Think about the last time you tried to get an AI tool to do something real and it fell short. You asked it to draft a contract, or read a messy spreadsheet, or write in your voice, and the result was close but not usable. So you filed it under "AI can't do this yet" and moved on.

That verdict is now out of date, and you have no way of knowing it from where you sit.

The model that failed you in March is not the model you are using today. Same login, same interface, different engine underneath. The thing you decided AI couldn't do might have quietly become the thing it does well while you weren't looking.

So do this. Keep a short list. Every time you ask AI to do something and it disappoints you, write down the task in one line. Not a project. Just "couldn't summarize the client call cleanly" or "botched the invoice math." Then once a quarter, spend twenty minutes running that list again from the top.

Some of it will still fail. That is fine, you learn where the real ceiling is. But two or three items will suddenly work, and each one is a task you were still doing by hand for no reason.

Here is where I want to slow down. The instinct when a new model drops is to go shopping. Switch tools, chase the benchmark, sign up for the thing everyone is posting about. Resist that. You almost never need the newest model on release day. What you need is to notice that the tools you already use got better, and to act on it.

The gap that costs you money is not the gap between your AI and the best AI. It is the gap between what your AI can do now and what you last checked it could do.

There is a second thing hiding in this release worth a sentence. Sonnet 5 is built to run agents cheaply, aimed at businesses using AI to actually do work, not just answer questions. Translation: the cost of pointing AI at a repetitive, boring task just dropped again. The math on automating something tedious gets a little more attractive every time one of these lands, whether or not you are watching.

So here is what the week actually leaves you with. Your opinion of what AI can do has a shelf life, and it is shorter than you think.

Most people form that opinion once, early, when the tool was at its worst, and carry it for years. They are running a company on a verdict they reached during the trial version.

The tools are moving faster than your memory of them. The only way to keep an accurate picture is to keep checking, because the version of AI you decided you didn't need may not be the version sitting in your account right now.