Claude Fable 5 vs GPT 5.5: same job, two results

The day Claude Fable 5 came out, I put it straight in the ring with GPT 5.5 in Codex.

Not on a coding puzzle. Not on a benchmark chart. On the kind of job my clients actually pay for: a premium website for a business that needs customers.

One model won on looks. The other one might win for your wallet. Let me show you both.

The test

Here's the setup. Same prompt, both models, side by side.

Fable 5 was available in chat and Cowork at the time, but not in the desktop app yet, so I ran it from the command line on high effort. GPT 5.5 ran in Codex on extra high thinking, because I like my stuff to be right.

The prompt, in short:

"Build a premium, high-converting single-page website for Conduit AI, an AI consulting business that helps electrical contractors win more jobs and run their office on autopilot. The buyer is a busy electrician or shop owner who is on the tools all day, hates admin, and is bleeding money to missed calls and slow follow-up. Sell the result, not the tech. Make it look expensive. Never a stock template. Dark and cinematic. Use real photos."

Why this test? Because I work with a lot of tradesmen. Plumbers, electricians, remodelers, the guys who put down the Sheetrock. They need something that separates them from everybody else. This is my benchmark. What can you actually do for a real business?

How Fable 5 worked

Fable listened. That was the first thing I noticed. It followed my direction, pulled real media, and on its own opened a browser to check that the site actually rendered, including the mobile view.

When it finished, it told me two things straight: the booking was a placeholder and the testimonials were samples. I respect that. It told me what was not real.

The site? Fire. Dark, cinematic, every section I asked for. It even had a little "find out what you missed" piece that hit the missed-calls pain right on the head. It made one mistake I caught: some of the numbers were too small. That is a quick fix.

To put it in perspective, the night before I spent about an hour with Opus 4.8 working on a page for my own site. Fable's one shot was better than that hour.

My score: 4.8 out of 5.

How GPT 5.5 worked

GPT 5.5 took its time. A long time. It generated its own visuals for the site, and I was sitting there thinking it was going to lose on speed alone.

Then it came from behind. When it opened, the headline read: "When the phone rings after hours, Conduit AI still picks up." That is selling the result, not the tech. Exactly what I asked for. It looked good.

And Codex itself has some things I love. Side conversations. Easy toggles. Claude can learn from that.

My score: 4.5 out of 5.

So who wins?

On looks, Claude Fable 5. No question.

But here's the part people skip. Inside the app it said Fable 5 was included in your plan limits until June 21st, and after that it switches to usage credits. Usage-based billing. On the other side, OpenAI had just ended the double usage they had been giving everybody.

So for the regular person, and for the clients I work with, GPT 5.5 might be the real winner. Because they are just as happy with what GPT 5.5 built, and it doesn't come with a meter that keeps spinning. That is Anthropic's problem to solve, not yours.

Me personally? For my own site, I might reach for Fable. For a lot of my client work, Codex does the job.

Don't get lost in the benchmarks

The day before, I was talking with another man who does AI consulting. He told me something I keep coming back to. People like us are maybe 1 to 5% of folks who even know about this stuff. Go out on the street and most people barely know what Claude is. When they think AI, they think ChatGPT.

So there are two ends. The people who know everything, who benchmark all day and never make anything. And the people who know nothing, who are getting passed by. Neither one gets an outcome.

You want to be in the middle. These are tools. I brush my hair to keep my waves. Same idea. A brush is a tool. Fable is a tool. Don't bow down to it. Use it to do something.

And the real promise of these tools is not "we're going to replace you." It's "we're going to buy back your time." You can always make more money. You can't make more time.

Look, I pay 200 dollars a month for this stuff. You best believe it better be making me money and making my life better.

What this means for you

Test the model on your real work, and look at what it will cost you to keep using it, not just what it scores. The best-looking result is not always the best choice for the business.

If you don't have time to run these tests yourself, that is what the RMDW Program is for. A mentor every week, working with you to put the right AI to work in your business, so you get the outcome without living on the benchmark charts.