Heidelberg AI News

Video · Tech With Tim ·

Claude Fable 5.2 is INSANE

In short

The video tests Claude Fable 5.2 by giving it one prompt and letting it independently build three apps overnight: a WebGL2 game, a programming language with an IDE, and a multiplayer whiteboard. The experiment shows impressive autonomous coding capacity, but also rapid usage consumption and potentially high costs for larger projects.

Key points

  • The three apps took 3 hours 45 minutes and produced about 14,000 lines of code.
  • Emberwake is a 3D glider game built with raw WebGL2 in about 1 hour.
  • Nocturn is a statically typed programming language, WebAssembly compiler, and IDE with 281 tests.
  • Converge is a multiplayer whiteboard with a hand-written WebSocket protocol, custom CRDT, and zero npm packages.
  • The builds used about 60 million tokens and 26% of the weekly Fable 5.2 limit.
  • API-equivalent usage for all three apps was estimated at $92.

Written by an AI model from the video’s captions and description.

I gave Claude Fable 5.2 one prompt, went to bed, and woke up to three finished apps. In this video I show you the exact prompt, what it built overnight, and what it actually cost in tokens and usage limits.

Watch on YouTube The opening of the article at youtube.com.

In this story

A tool or model opens in the Heidelberg AI Curriculum: what it does, what it costs, how to start. A repo opens on GitHub.

Related stories

All stories

VideoWes Roth

Claude JUST got MYTHOS...

The video argues that Claude is evolving into an AI operating-system layer that coordinates office tools, meetings, data, software, and third-party apps rather than replacing each app directly. It also explains how eligible independent researchers, open-source developers, and vulnerability hunters may apply for Mythos 5.2 through Anthropic’s Cyber Verification Program, while challenging the interpretation of the MIT study claiming 95% of enterprise AI projects produce no ROI.