Fable 5

The most capable model in my sessions, and the one that types the least. It holds the plan, writes the specs, reviews the diffs, judges the edge cases, and owns the final verification. When a cheaper session hits a wall twice, the escalation is a consult here, not a third retry.

Plans and judgments compound; keystrokes don't. Spending the strongest model on deciding — and cheaper ones on typing — puts the quality where it multiplies.

Its time goes where reasoning is genuinely hard: decomposition, architecture calls, ambiguous synthesis, the last read before ship. Everything mechanical gets pushed down the ladder so this tier's attention stays on judgment.

Runs in my coding sessions as the planner and reviewer tier. The transferable practice: give your strongest model the review-and-decide role, not the typing.