Heidelberg AI News

Video · Fireship ·

Did Anthropic just save its $1 trillion IPO?

In short

The video examines Anthropic’s Fable and Mythos 5.2 models, released during a difficult IPO roadshow, and evaluates their speed, cost, autonomous coding, hardware design, and 3D generation capabilities. The demonstrations suggest major gains for long-running, multi-agent software and engineering tasks, though the models still have limitations and are not universally best.

Key points

  • Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, down from version 5.1.
  • A swarm of 1,000 Fable agents built a bootable operating system from an empty folder in eight days.
  • Fable agents formalized Fermat’s Last Theorem in Lean in 30 hours, while Opus 5.5 reached 26% before crashing.
  • Fable designed a voice assistant hub, USB speakerphone, and handheld virtual pet from a one-paragraph prompt.
  • Fable beat GPT-6 Astra on Anthropic’s listed benchmarks but lost to Gemini 4 Carbon in business automation.
  • A 3D horse generated with Three.js used over 120,000 vertices, 38 bones, and inverse kinematics, earning a 7.77 Throughbench score.

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

Anthropic just dropped Fable and Mythos 5.2 in the middle of a rocky IPO roadshow. Let's dive into what they can do 🐎.

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

VideoRiley Brown

Claude Just Changed EVERYTHING (Here’s What’s New)

The video surveys new AI-agent capabilities across Claude, Gemini, GrokBot, and Dots, with emphasis on Claude’s workspace artifacts, meeting participation, model tradeoffs, and computer control. It argues that agents are becoming operational coworkers, but payment restrictions and platforms blocking agents remain major barriers to running businesses with them.