I spent all day running GPT-5.6 Sol against Claude Fable 5, side by side, on real work instead of just staring at benchmarks. I put them head to head in Codex and Claude Code on browser games, interactive websites, and open-ended builds, then ran a bunch of quick one-off tasks over the API to see how they compare on speed, cost, and reliability. My take: Fable is the better manager and the more creative, capable model, while Sol is a really good worker that ships fast and costs a fraction of the price. Watch to see where each one wins and how I decide which model to reach for.

Fable 5 is going back behind subscriptions at some point, so I've been focused on keeping its process instead of its intelligence. In this video I walk through how to extract the way Fable works into a skill that makes Opus 4.8 feel elevated, how to actually use effort levels, and how to set up a simple model routing table so cheaper models handle the work they're capable of.