I was a Claude hater. Fable 5.2 changed my mind.

Summary

The host reviews Anthropic’s Fable 5.2, a model that changed her opinion of Claude after earlier versions felt capable but frustrating to use. She describes Fable 5.2 as fast, creative, and pleasant to work with, while emphasizing that it does not replace Opus 5.5. Opus is positioned for everyday coding and knowledge work, while Fable is better suited to tasks requiring greater complexity, scale, or creativity. Claude Mythos 5.2 is described as the same model, restricted to cybersecurity and life-sciences uses.

Main Findings and Demonstrations

• Performance and coding: The host cites strong benchmark results and says the model handled complex bugs, optimization, architecture, migrations, and long-running tasks well. She used it to reorganize a messy product-intelligence pipeline for ChatPRD, reporting strong architectural reasoning and code quality. The migration was still in progress, however, so the result was not fully validated.
• Video and animation: Given a tweet and a brand book, Fable created a detailed animated marketing video without a supplied storyboard or much creative direction. The host especially liked its animation of existing static “invader” characters and plans to include this task in a future How I AI evaluation benchmark.
• Geometry and visual reasoning: The host supplied low-quality photos of furniture parts, including labels in Chinese, without a manual. Fable produced diagrams and assembly steps that she says worked when followed.
• Game creation: From a short prompt and the company’s existing character art, Fable built a Space Invaders-style bug-defense game and a humorous trailer. The game connects software-engineering concepts—such as agents, bugs reaching production, and token costs—to its mechanics.
• 3D interactive app: The host’s standout project was a rendered typewriter with keyboard input, multiple settings and scenes, sound and music, writing timers, and the ability to print and save text. She describes it as a creative project she had not previously imagined she could build.

Limitations and Caveats

• Web design: The host found Fable’s independent frontend and web-design choices poor. It performed better when given a strong brand book or style guide.
• Image generation: She often turned to another image-generation tool for visuals, saying Fable was not strong enough at image creation for her needs.
• Evaluation status: The review is based on a few days of hands-on use. The host plans a fuller How I AI comparison with other models; the examples and benchmark claims in this episode are not presented as a completed independent evaluation.
• The host also mentions cybersecurity and biodefense safeguards, as well as protections against model distillation, but does not explain their details.

Fable 5.2 is expected to be available through the API and cloud platforms, including AWS, Google Cloud, and Microsoft Azure, as well as Anthropic’s cloud applications and Claude Code. A faster mode is also mentioned.

Main Speakers and Sources

• Claire, host of How I AI, provides the review and hands-on demonstrations.
• Anthropic is cited for the model’s positioning and benchmark claims.
• Guillermo from Vercel is mentioned as the inspiration for the typewriter’s integrated record-player concept.
• Jamie Gannon is credited with creating the original static character sprites used in the demonstrations.