When NOT to use Fable 5

The habit that quietly drains your credits

July 05, 2026

In partnership with

I keep watching people grab their most powerful model for the most basic jobs, then act shocked when their credits evaporate. Reading a PDF. Summarizing a webpage. Cleaning up a draft. All handed to the heaviest, most expensive model in the drawer, like using a forklift to carry a coffee cup.

So when I found a post from an AI pro laying out exactly when not to reach for Fable 5, I stopped and read the whole thing twice. It flips the instinct most of us run on: "just use the smartest model." The list is short and blunt, and it made me rethink how I route my own work.

The core idea is simple. Raw power is not the same as the right fit. Fable 5 is a heavy, highly automated model. Point it at simple work and you are paying premium rates to do something a lighter model does just as well for a fraction of the cost.

Blu Dot surpasses 2,000% ROAS with self-serve CTV ads

Home furniture brand Blu Dot blew up on CTV with help from Roku Ads Manager. Here’s how:

After a test campaign reached 211,000 households and achieved 1,010% ROAS, the brand went all in to promote its annual sales event. It removed age and income constraints to expand reach and shifted budget to custom audiences and retargeting, where intent was strongest.

The results speak for themselves. As Blu Dot increased their investment by 10x, ROAS jumped to 2,308% and more page-view conversions surpassed 50,000.

“For CTV campaigns, Roku has been a top performer,” said Claire Folkestad, Paid Media Strategist, Blu Dot. “Comping to our other platforms, we have seen really strong ROAS… and highly efficient CPMs, lower than any other CTV partner we've worked with.”

Using Roku Ads Manager, the campaign moved from a pilot to a permanent performance engine for the brand.

*Ad

Five jobs Fable 5 is wrong for

Here is the breakdown from the original post, plus a little context on why each one lands.

Reading large documents or ebooks. Long-form reading is grunt work. You do not need a heavyweight model to chew through pages of text, and doing it anyway quietly bleeds your budget.

Analyzing hundreds of images or videos. Bulk visual processing is about volume, not deep reasoning. Classic overkill. A lighter model runs the same batch without the premium price.

Writing prompts or stories. Creative work and prompt drafting do not need the model's heaviest machinery. Lighter models handle this cleanly, and often faster.

Web-based research. Pulling and summarizing info off the web is routine. Spending premium credits here is money left on the table.

Saying "everything." This one is sneaky, and it is the best point in the whole post. It gets its own section.

The "everything" trap

Here is the detail that surprised me. The author warns against dropping words like "everything" or "perfect" into your prompt. Those vague, all-encompassing words trigger the model's automated nature to go do far more work than you actually asked for.

How much more? According to the post, token consumption can run 5 to 6 times higher than Opus 4.8, purely because of how automated the model is.

Read that again. A couple of lazy words in your prompt can multiply your cost five or six times over. That is the kind of hidden leak that does not sting once, it stings every single time you run the task, all day, every day.

The fix is almost insultingly simple. Say what you want, specifically. "Everything" and "perfect" feel thorough, but to an automated model they read as "go maximize effort with no ceiling." Give it a ceiling.

See what enterprise-ready AI support looks like

How are leading teams getting AI support to work? We're breaking down the playbook, live July 9.

*Ad

So when should you actually use it?

The person behind this is not saying the model is useless. Far from it. Their advice is to save it for a genuinely difficult problem that other models simply cannot crack. Hand it something hard, and you might be surprised by what it pulls off.

Use it as a specialist, not a default. That reframe alone is worth the price of admission. The heavy model is not your everyday driver. It is the expert you call in when the normal team is stuck.

A real example: the NotebookLM fix

The post shares a perfect case of what not to waste Fable 5 on. If you have made a video with NotebookLM, you know it can spit out a bunch of static, frozen frames. Annoying. The instinct is to throw your biggest model at fixing it. Do not.

Instead, hand the job to Claude Code with a clear instruction. Here is the exact prompt they shared:

Extract the static frames from this video and the duration of each one. Use the [MCP name] MCP to transform every static frame into a short video, then assemble them into a new video based on the same timestamps. [Describe your caption style if needed.]

You swap in the name of your video-generation MCP where it says [MCP name], and add your caption style if you want one. The whole thing runs nicely on Opus or Sonnet. No premium heavyweight required at all.

That is the pattern in miniature. A task that looks intimidating is really just routine work with a few steps. Route it to a mid-tier model with a clear prompt and it gets done for a fraction of the cost.

What’s next is almost here.

On July 16th at 1PM ET, beehiiv is going live with a look at the future of publishing, audience growth, and digital business.

What started as a newsletter platform has evolved into something much bigger: a place where creators and brands can grow, monetize, and own their audiences without stitching together half the internet to make it work.

The next chapter starts live at the Summer Release Event. 

Join us to see what’s coming next.

*Ad

Why this actually matters

Underneath all of it is one lesson: match the tool to the task. We have been trained to think "smartest model wins," but the sharper point is that the smartest model is usually the most expensive way to do simple work.

The line that stuck with me: do not waste your credits on Fable 5 for tasks like this, even if you are rich. It is not about whether you can afford it. It is about not being wasteful when a lighter model does the same job better and cheaper. Being deliberate about which model handles which job is how you keep your credits alive long enough to spend them where they count.

A few moves you can make today:

Strip vague words like "everything" and "perfect" out of your prompts before you hit enter. That alone can cut runaway token costs.

Reserve your heaviest model for problems that genuinely stump everything else. Specialist, not default.

Route routine reading, research, and bulk processing to lighter, cheaper models. They handle it fine.

For NotebookLM's static-frame problem, let Claude Code plus a video MCP run it on Opus or Sonnet.

The part I keep coming back to

The reason this hit me is that I have been the person burning premium credits on a webpage summary and never noticing. Not because I needed the power, but because reaching for the biggest model felt like the safe, serious choice. It was neither. It was just expensive.

The unlock is not a smarter model. It is a smarter habit. Before you send the next prompt, ask one question: is this actually hard, or does it just feel important? If it is routine, route it down. Save the heavy model for the problem that earns it.

So tonight, open the last three tasks you handed your most powerful model. Be honest about which ones a lighter model could have finished just as well. Move those over, and watch what happens to your credits by the end of the week.