Did Anthropic just save its $1 trillion IPO?

Summary

The video reviews Anthropic’s Fable and Mythos 5.2 models, released during a difficult IPO roadshow. The narrator says investor concerns about data centers and a broad AI-stock selloff had reportedly pulled Anthropic’s expected valuation down from $2 trillion to $1 trillion. The release is presented as a way to refocus attention on the company’s technology.

Models and pricing

Fable and Mythos are described as the same underlying model, with Mythos unrestricted and aimed at large organizations, while Fable has safety guardrails. Pricing falls to $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for version 5.1. Anthropic reportedly recommends Opus 5.5 for most work, reserving Fable for tasks that need it.

Software and math demonstrations

In one test, Fable agents built an operating system from an empty folder over eight days. The result reportedly booted on a real laptop with a kernel and drivers.

A swarm of 1,000 agents also completed a formalized proof of Fermat’s Last Theorem in Lean in 30 hours. The proof was said to be seven times shorter than an earlier attempt, while Opus 5.5 failed to complete the same setup.

Hardware design

Working with circuit-board company Diode, Fable reportedly turned a one-paragraph prompt into designs for a voice-assistant hub, a USB speakerphone, and a handheld virtual pet.

The video cites benchmark results favoring Fable over GPT-6 Astra on Anthropic-listed measures, while Gemini 4 Carbon reportedly led in business automation. It also notes that Fable more than doubled Astra on a robotics benchmark run by Anthropic.

Interactive 3D horse test

The narrator used Claude Code to request a realistic, interactive horse in Three.js. Fable spent 43 minutes creating it from signed distance fields, then reviewed and revised its own first attempt.

The finished browser-based horse features mouse tracking, keyboard movement, orbit controls, and animated gaits. The narrator praises its appearance and smooth movement, while noting flaws including a lopsided mane, awkward stopping, and eyes that are not fully open.

Its model reportedly has more than 120,000 vertices and 38 bones, with inverse kinematics for hoof placement. The narrator awards it 7.77 on “Thoroughbench,” calling it the best horse produced in that test so far. The horse is available to try at typescript.dev.

Sponsor segment

The video promotes Mux’s video API and Mux Robots. It says uploads can automatically generate transcripts, storyboards, thumbnails, and clips. Robots can run AI workflows—such as finding key moments or selecting thumbnails—without users hosting models or maintaining a pipeline.

Workflows can be set to run on new uploads. The sponsor segment also mentions a free tier and a $50 credit offer.

Guides or tutorials

There is no full tutorial. The horse demonstration shows a prompt-based coding workflow in Claude Code, but is presented primarily as a model test and review.

Main speakers and sources

The video is narrated by Fireship’s host. Model capabilities, pricing, and benchmark results are attributed to Anthropic or its demonstrations. The video also includes a sponsored segment about Mux.