Okay, Anthropic—what exactly are you testing behind the curtain?

While experimenting with Fable 5.0 Medium, I entered a single prompt asking it to create a Titanic animation. What came back wasn’t some basic floating ship demo. The model generated an impressive cinematic experience using Three.js, complete with lighting, atmospheric effects and an ocean that appears to use its own custom-generated shader.

And here’s where things get seriously interesting: the entire creation weighs in at only 91KB.

Medium Just Beat Max—And It Wasn’t Close

To put that result to the test, I gave Fable 5 Max the same prompt. You would naturally expect the higher-effort Max setting to produce the superior creation—but that isn’t what happened.

The Fable 5 Max rendition was more than 100KB, yet it still didn’t look as polished or visually impressive as the 91KB version produced by Medium.

So let’s get this straight: Medium generated the better-looking animation, seemingly created a specialized shader for the ocean and delivered everything with considerably less code?

Yeah, that raised an eyebrow.

Was This Really Fable 5.0?

The obvious question is whether this was simply an unusually strong Fable 5.0 run—or whether Anthropic quietly handed my account something newer.

According to rumors, Anthropic may be conducting a limited “greyscale” test of Fable 5.1 across a small subset of accounts.

The theory is that selected users could receive the upgraded model behind the scenes while the interface continues to display the regular Fable 5 name. That means two users could choose the same model, enter the same prompt and receive noticeably different levels of performance.

Sound familiar?

The 91KB Clue

File size alone cannot prove which model generated something. However, the combination of visual quality, code efficiency and shader work makes this particular result difficult to ignore.

• Generated from one prompt
• Built with Three.js
• Includes an apparently custom ocean shader
• Only 91KB in size
• Looks better than the 100KB-plus Fable 5 Max attempt

Could normal output variation explain the difference? Absolutely. AI models can produce dramatically different results from one run to another, even when the same prompt and settings are used.

Anthropic could also be testing an improved system prompt, a new inference configuration or some other optimization that has nothing to do with Fable 5.1.

But if the rumors are accurate—and Anthropic really is slipping an experimental upgrade to selected users—then this is exactly the kind of unexplained performance jump we might expect to see.

Did I Catch One of Anthropic’s Secret Tests?

There’s no model ID, official announcement or hidden label I can point to as definitive proof. Anthropic has not publicly confirmed that Fable 5.1 is being tested.

Still, based on what this Medium run produced compared with Fable 5 Max using the same prompt, it certainly looks possible that I caught one of those rumored tests.

Either Fable 5.0 Medium suddenly had the run of its life—or Anthropic’s next model just sailed onto my screen disguised as the old one.

Watch the complete Titanic animation here:
https://www.youtube.com/watch?v=KjB8rLMyJ00