When to reach for Fable 5

Fable 5 is the most capable model we've put our hands on — a million tokens of context and reasoning that thinks before it answers by default. The temptation is to point it at everything.

We don't. A smaller model handles most of the work just as well — and faster: summarising a meeting transcript, generating the first draft of a routine contract section, turning a support thread into a structured incident report. For those tasks, Fable 5 is not just expensive; it is slower. The milliseconds matter when something runs hundreds of times a day.

We reach for Fable 5 when the task requires holding complexity across a long horizon without losing the thread: redesigning a multi-step approval workflow where every edge case interacts with every other; writing the architecture for an agent that has to reason about its own failure modes; or reading a dense legal document in context with a client's existing contracts and surfacing only the conflicts. These are not hard because they are technical. They are hard because they require sustained coherence — keeping the whole thing in view while working on any one part.

The non-obvious thing is this: a smaller model will often give you an answer that is locally correct. It does what you asked. But it may not have noticed that what you asked was slightly wrong — that the framing had a flaw, that one assumption downstream would cause a problem. Fable 5, used well, pushes back on the framing. That is the real reason to reach for it.

"With AI" means pairing judgement with generation — and judgement includes knowing when the expensive model earns its keep. A sledgehammer is a wonderful tool. You still don't use it to hang a picture.