I tested Claude Sonnet 5.5 on a real client

Anthropic dropped Claude Sonnet 5.5 right after Opus 5.5. They are working on me, man. Trying to turn me back into a fanboy.

But I don't care what a model scores on a chart if it can't do the work my team and I actually get paid for. So I did not test it on a puzzle. I tested it on a real client.

Here's what the numbers say, what it built, and what I think it means for which AI you should be using.

What the benchmarks say

Anthropic calls it a clear upgrade over Sonnet 5. They say it runs about 30% faster and costs about 30% less for most work.

On their chart, a few things jumped out at me:

• Agentic coding (Terminal Bench 4): 70.6%, which is actually higher than Opus 5.5.
• Agentic coding in Cursor: 55, with Opus 5.5 at 57.
• Computer use: 80.1%, with Opus 5.5 at 81.8%.
• Knowledge work: about two points under Opus 5.5.
• Multidisciplinary reasoning: about three points behind Opus 5.5.

They also say it has the best cost per attempt of any model out right now. Apparently. I'm on a subscription, so I don't pay API rates and I can't check that one myself.

So on paper, it is close to Opus for less money. Fine. Let's see it work.

The real test: a trades business in Atlanta

A few days before, I talked with a gentleman who runs a trades business here in Atlanta. Sheetrock, plumbing, electrical. He knows people and brings them together. That is his business.

What he needs is a website that shows off what his team does and lets customers book.

So I spoke this prompt to Sonnet 5.5, and I told it not to use any of the skills we already have set up. Use your imagination. Here's the gist of what I said:

"I need a premium-looking website for a tradesman company. Their specialty is Sheetrock, and they also do electrical and plumbing. It needs to be dynamic and eye-catching, something that makes people want to buy. I need a hero section, a call to action, and a pricing section, but the pricing needs to be smart, with starting-at prices. People need to be able to book and have that sent to the business. And it has to look amazing. In his field, he needs to stand out."

What it built

Straight walls, live wires, zero leaks. That was the headline it came up with, and I liked it.

The site was clean, professional and simple. I can tell when something is Sonnet, but this one stood out because it had elements a customer can actually play with. On the Sheetrock section, it shows an example quote priced per square foot and walks the customer toward booking the job.

Was it perfect? No. It did not one-shot as much as Opus 5.5 does. I would have to ask it for two or three more things. But that's fine. Two or three more asks and the job is done, on a model that is faster and cheaper. That is a good trade.

The model is the smallest part

Here's what I want you to hear. A first draft from a model is not a finished website. That is not how business works, and it is not how customers end up happy.

After a draft like this, the RMDW team goes back in. We handle how bookings get emailed to the owner. We look at hooking it up so it can take calls and book more jobs. We add real photos of his work once he sends them. Our principal engineer maintains the code, keeps it secure and makes sure we can track who visits and what they click. Our branding person makes sure the logo looks right and draws people in.

And we stick with the customer. We don't make the site and walk away.

Then there's privacy. Say he wants a built-in app on that website that tracks his customers, bookings, contact info and past jobs. That is information he would want kept private. That is where we would run Jah 6.0, our own model, as the engine behind that part of the site. Use the public model for the public work. Use our own for the private work.

Which AI should you use?

My honest take: 99% of people are not doing anything that needs Opus 5.5, Fable or Sonnet 5.5. The models are becoming a commodity. That's why I keep telling you to go hybrid and keep options, not box yourself into one.

But when you do real work across websites, apps, databases and logos like we do, you learn that certain models are better at certain things. And affordability matters right alongside privacy and security. Sonnet 5.5 is one we will definitely be using at RMDW.

What this means for you

If you are a business owner, the question is not which model wins this week. The question is who is going to take the draft and finish it: the booking, the follow-up, the security, the upkeep.

That is what custom builds from the RMDW team are for. Websites that book jobs, AI that answers your calls, and the apps your business runs on, built and then looked after.