1. News
  2. AI models & agents
  3. Anthropic · Models
Anthropic · Models2 min read

Anthropic shows what its new Claude Fable 5.2 model can build from code

Anthropic's Claude account showed seven early demos of Fable 5.2, from a 3D cell to a chalkboard. They also hint at how fast paid usage limits can run out.

By the Dashline deskSource: Pasquale Pillitteri
Anthropic shows what its new Claude Fable 5.2 model can build from code
Photo: Alphaaloy, CC BY-SA 4.0

Anthropic, the company behind the Claude AI assistant, has shown what its new Fable 5.2 model can do. A thread on Claude's X account on November 10 featured seven demos, including a 3D model of a cell built entirely from code, one organelle at a time. Organelles are the small parts inside a cell, like the nucleus.

What was shown

The cell video runs 2 minutes and 14 seconds. It was the most viewed post, with about 90,000 views. Kevin Ngo, a creative who codes with AI, built it. The other demos include:

  • A podcast on the Egyptian pyramids, with images drawn in code (2 minutes 7 seconds).
  • A chalkboard that explains any answer one step at a time, by Tyler Nishida.
  • A prototype that lets you watch subagents, helper copies of the model, work in a project.
  • A human body in motion, drawn from one prompt by three Claude models.

The body post does not name the three models or say which one is Fable 5.2. The origami demo is unsigned too.

What it costs

Fable 5.2 came out on November 9 on all paid plans: Pro, Max, Team and Enterprise. It is also available through the API under the name claude-fable-5-2. The API price is 8 dollars per million input tokens and 40 for output. A token is a unit of text, roughly three quarters of a word in English. On Pro, it is included within usage limits until November 16, up to 50% of the weekly limit.

Long jobs like these can eat those limits quickly. Using numbers from a creator named Theo, the 200-dollar plan runs out much sooner with Fable 5.2 than with Opus 5.5. That is about 3,000 dollars of usage a month against 15,000.

How to read the demos

This is a showcase picked by the maker's own account, so it is marketing as much as proof. There are no measurable comparisons in the thread. Still, one pattern shows up. Most demos start from code the model writes, runs and checks, then fixes. The value may lie less in the final image and more in that check-and-retry loop.

What is not known yet

The prompts behind the experiments have not been published, so nobody can easily repeat them. It is also unclear whether the cell project will be shared as open source. Ngo's earlier cell diagram, made with Opus 5.5, was. Until the code or prompts appear, the demos show what is possible, not what an everyday user will get on a first try.

The takeaway

Fable 5.2 demos look impressive, but long jobs can drain paid usage limits fast, so plan your prompts before you start.

More articles

You should really read these.

Next step

Let’s talk