As I use Claude Opus 5.5 more, I find it capable enough to be the orchestrator, with xhigh as the effort level.
But what about Fable 5.1? Well, I find that Claude Code has an /advisor mode, which allows me to specify any model as the advisor, and the current agent will consult the advisor when facing tough questions, especially if my prompt explicitly says “consult the advisor when you need to make decisions about design or implementation.” To my surprise, the Opus 5.5 orchestrator would consult Fable 5.1 quite often using such an /advisor mode, but not too often such that tokens are not wasted unnecessarily.
With the arrival of Sonnet 5.5, I now ask my orchestrator to use Sonnet 5.5 for its explorer subagents, and Opus 5.5 for its implementer and reviewer subagents. Fable 5.1, of course, will always be the model for its writer subagents. I am using this workflow for some pretty ambitious projects (such as improving the Days network simulator), and it seems to be working quite well so far.