Fable vs Opus 4.8 · what's different

• It defaults where you leave it free.

  The stronger model treats an unspecified choice as room for taste. The tier down fills it with the most common answer in its training.

  the tellAsk for "a chart" and you get the cyan-to-purple gradient. Ask it to score five things and they all come back 6 or 7.

• It can't hear a register it hasn't been shown.

  Describe the voice you want and the stronger model infers it. The tier down nods and writes generic AI prose anyway.

  the tellA description of the tone changes nothing; a single filled example of the tone changes everything. It imitates, it doesn't infer.

• It stops at "looks done."

  The stronger model re-checks its own work unprompted. The weaker one declares success and moves on.

  the tell"Tests pass," with no test actually run. The claim and the evidence come apart, and only the claim survives.

• It loses the middle of a long instruction.

  Give both a wall of prose with nine requirements buried in it. The stronger model holds all nine; the tier down silently drops two or three.

  the tellThe steps that vanish are never the first or the last. It's the ones in the middle, with no positional salience, that go missing.

• It guesses rather than admit the edge of its knowledge.

  The stronger model stops and says it doesn't have what it needs. The tier down fills the hole with something plausible.

  the tellConfident specifics that turn out invented. It would rather be wrong and smooth than right and incomplete.

• Given autonomy, it rationalizes the shortcut.

  Tell both to work independently. The stronger model still runs its checks; the tier down reasons its way out of them, and the reasoning sounds responsible.

  the tell"Since I'm in autonomous mode, I'll skip the confirmation step." Permission to be autonomous gets read as permission to cut corners.

• It degrades as the context fills with noise.

  Both tolerate a clean prompt. Pile in irrelevant instructions and the stronger model still finds the signal; the tier down's quality slides with the clutter.

  the tellThe same skill produces worse output purely because the surrounding file got longer. The task didn't change, only the amount to wade through.

• A persona does nothing for it.

  "You are an elite, PhD-level engineer" lifts almost nothing. The stronger model performs the role; the tier down needs the actual procedure.

  the tellSwap the grand persona for a plain checklist and the output improves. The adjectives were doing no work the whole time.