GPT-5.6 Sol is crazy for marketing work

GPT-5.6 Sol is crazy for marketing work.

I tested GPT-5.6 Sol against Claude Fable 5 on four marketing tasks.

Actual work I care about:

You rebuild a website concept. 
You package YouTube thumbnails. 
You analyze growth opportunities for a large business. 
You find clips from a recording that can become short form, mid form, and long form content.

Not anymore.

The gap is showing up in the place marketers actually care about: finishing.

The website test showed me who could create a usable world

The first prompt was simple:

"Install the skill and build singlebrain.com into a world."

That was basically it.

Sol took the existing Single Brain concept, revenue agents that know the client, and turned it into an interactive world. It created the idea that a company has one main brain, then team brains, then individual brains underneath it.

I did not feed it that hierarchy.

It inferred it.

The first version still needed work. But I had something to react to. I could scroll through it, see the world concept, and decide where to take it next.

The run used about 467 credits and around 766,000 tokens over roughly 34 minutes.

One basic prompt got me a prototype I could critique.

That changes the creative loop.

The thumbnail test exposed the real marketing bottleneck

The second test was YouTube packaging.

I gave Sol a transcript from a recent video and asked it to look at my Leveling Up thumbnails to determine the channel's thumbnail identity. Then I asked for three thumbnail and headline options based on the transcript.

It worked for about 6 minutes.

The first output was not perfect, but it looked like my channel.

Then I gave it feedback: reimagine this one, try this direction, use the right visual mark.

Pretty quickly, I had usable directions for a Sol 5.6 versus Fable 5 thumbnail.

Thumbnail packaging drives distribution. If the thumbnail and headline are wrong, the video underperforms before the content gets a chance.

The old version of AI gave you generic thumbnail ideas.

This version gets closer to a repeatable creative system.

The next move is obvious: turn the thumbnail identity into a skill so every future video starts from the same taste layer instead of a blank prompt.

Fat skills, thin harnesses.

The business analysis test proved speed is not the whole game

I also tested business analysis.

The prompt was to advise senior leadership at Amazon on the three incremental growth initiatives that would create the most enterprise value over the next three years.

Sol worked for about 9 minutes and came back with a cleaner answer:

AWS AI utility. 
Amazon ads. 
Amazon supply chain services.

It also gave allocation logic: roughly 70 to 80% toward AWS AI, 10 to 15% toward ads, and 10 to 15% toward supply chain services.

That is the kind of output you can debate.

You may disagree with the percentages, but at least it gives you a point of view, a ranking, and a capital allocation frame.

Fable 5 leaned toward the grocery network. I think that was a miss.

That is the difference between an AI that summarizes and an AI that helps an operator make a decision.

The content clipping test is where I saw the biggest marketing unlock

The last test was content repurposing.

I asked the model to clip up a webinar for social media.

Short form: 60 to 90 seconds. 
Mid form: 2 to 5 minutes. 
Long form: around 10 minutes with highlight montages and cliffhangers.

This is the workflow every content team wants to scale.

The model started identifying moments and creating assets. Some clips still needed audio treatment. Some were choppy. But a few moments were usable.

One clip had a hook around old management styles breaking in the AI era. Another had a broader story about homebuilding technology taking longer than expected. Another turned into a leverage analogy.

That is the unlock.

Finding moments used to be a taste problem.

You needed someone to watch the recording, understand the market, find the hook, know what has reach potential, and protect the brand.

Now the human's job changes.

You do not start by finding the moments.

You start by judging the moments.

That is a much better use of human taste.

My read: Sol is winning early marketing workflows

I still like Claude Fable 5. In some design moments, it created concepts I liked better.

But for these marketing tasks, Sol 5.6 won because it got further with less back and forth.

That is the bar now.

Not which model sounds smarter.

Which model finishes more of the work.

For operators, this means the AI stack is moving from assistant to production layer. Websites, thumbnails, growth planning, creative clipping, and repurposing are just the start.

Watch the full breakdown here: https://www.youtube.com/watch?v=wZKe9aL8iSs

To testing new models,

Eric Siu