The AI you can actually use just got a big upgrade
Last week the most powerful AI on earth launched behind a velvet rope. This week the opposite happened. Claude Sonnet 5 is out, it is for everyone, and the headline feature is one a normal business can really use.
There are two kinds of AI launch. The kind you read about and cannot touch, and the kind that quietly turns up in the tools you already use.
Last week was the first kind, when OpenAI put its new flagship behind a locked door. This week is the second kind, and for a real business it matters far more.
What happened
Anthropic has released Claude Sonnet 5, its new everyday model, and it is now becoming the default across its apps and tools. No waitlist, no partner programme, no velvet rope. If you use Claude, you are likely using it already.
Two things stand out. It can hold about a million tokens in its head at once, roughly 750,000 words. And it launched with a price cut, $2 in and $10 out per million tokens on the API until the end of August, cheaper than the model it replaces.
Why the memory is the real story
Most AI disappointments in a small business come down to the same thing. The model could not see enough of your world. You fed it one email when the answer lived across forty. You pasted in a snippet of your price list and it guessed the rest.
A million-token memory changes the shape of the job. That is a year of customer emails, every quote you sent, and your full price list, in one conversation. You stop feeding it crumbs and start handing it the whole situation, then asking real questions. Which quotes did we win, and what did the winning ones have in common? Which customers went quiet this spring?
One honest caveat on that price cut. Sonnet 5 uses a new tokeniser, the thing that chops your text into the pieces the model reads, and it reportedly produces around 30 percent more tokens for the same text. So the per-token price fell, but each job uses more tokens, and the real saving is smaller than the headline. That is not a scandal, it is just how launch pricing works, and it is exactly the kind of detail worth checking before you move a workload. The launch rate runs to 31 August, then it steps up to $3 and $15.
The part worth noticing
Put the two launches side by side and the lesson writes itself. The frontier model you cannot touch will not change your Tuesday. The workhorse model that just got a bigger memory and a lower price will, if you point it at the right job.
Because the bottleneck has not moved. A model that can hold your whole business in its head still needs someone to decide what job it does, what good looks like, what it must never do, and where its work lands. Bigger memory raises the ceiling. Design is still what gets you to it.
So what should you do this week
Think of the one AI experiment you gave up on because the tool kept forgetting, or because you could only ever show it a fragment of the picture. That is the one to dust off. Give the new model the whole picture and see what changes. Keep a human check on the output, same as ever.
And if you never had that experiment, here is a simple first one. Gather a year of quotes, wins and losses included, hand them over in one go, and ask what the winning ones had in common. Twenty minutes, and you will learn something about your business and about what these tools can now do.
Where we come in
This is what we do at Creative Sauce AI. We take a capable, available model, and this week that model got noticeably better, then we design the system around it that makes it safe to lean on for a real job in your business. The launches make the headlines. The design gets the results.
If you are wondering what a model that can hold your whole business in its head could do for you, that is exactly the conversation we love to have.
Book a quick chat →Related: The most powerful AI just launched. You can't use it, and you don't need to.
Common questions
What is Claude Sonnet 5?
It is Anthropic's newest everyday model, released at the end of June 2026 and now rolling out as the default across its apps and tools. Unlike recent frontier launches it is available to everyone, with a one million token context window and launch pricing of $2 per million input tokens and $10 per million output tokens until the end of August.
What does a one million token context window actually mean?
Roughly, the model can hold about 750,000 words in its head at once. In practice that is a year of customer emails, every quote you have sent, and your full price list, all in one conversation. You stop feeding it snippets and start handing it the whole situation.
How much does Claude Sonnet 5 cost?
Launch pricing on the API is $2 per million input tokens and $10 per million output tokens until 31 August 2026, then $3 and $15 after that. One honest caveat: it uses a new tokeniser that reportedly produces around 30 percent more tokens for the same text, so the real saving is smaller than the headline. In the apps, it is simply included in existing plans.
Should I switch my business to Claude Sonnet 5?
If a job in your business keeps failing because the AI forgets or you can only feed it fragments, this upgrade is worth testing on exactly that job. If your setup already works, there is no need to rush. The model is rarely the bottleneck. The design around it, the clear job, the checks, the handover, is what makes AI reliable.