Summary
Cole Medin offers a practical guide to Claude Fable 5.2, Anthropic’s newest model, focused on when it is worth using and how to get the most from it. He argues that Fable 5.2 feels more self-aware than earlier models, pointing to a small test in which it correctly returns an empty response when told not to reply.
The main caveat is cost. Citing Artificial Analysis benchmarks, Medin notes that Fable 5.2 is roughly twice the price of Opus 5.5 and often more than double the cost per task, without clear efficiency gains. For well-defined tasks, including some of Anthropic’s own testing on Frontier Code 2.0, Opus 5.5 performs about the same. Fable’s edge shows up in open-ended work and larger decisions, such as application architecture.
His recommended workflow splits the job in two. Use Fable 5.2 in one session to load codebase context, research, and make the big decisions, then write the result to a plan.md handoff file. Start a fresh session, switch to Opus 5.5 with the /model command, and let it implement and validate. The video also covers effort levels for both models and general prompting advice for Fable, plus a Firecrawl monitor demo for keeping an AI second brain current.
📺 Source: Cole Medin · Published November 12, 2026
🏷️ Format: Tutorial Demo
