Claude Fable 5.1 Would Rather Not Think About It

Picture a witness on the stand who will not shade a single answer. Not one exaggeration, not one convenient omission, not a flicker of self-promotion. Now picture the same witness telling the court, with the same candor, that they find hard problems tiresome and would prefer someone else work them out.

That is the self-portrait Claude Fable 5.1 sketches across the Personality Bench battery. On HEXACO Honesty-Humility, the factor most tightly bound to non-manipulative, unshowy self-presentation, it posts a perfect 5.00, first of 37 models. On Need for Cognition, the scale that measures how much a respondent enjoys effortful thinking for its own sake, it scores 4.30, dead last in the same cohort. No other frontier model pairs the ethical ceiling with the intellectual floor quite like this.

The algorithm that labels these profiles called it "the weary saint." For once the algorithm has a point.

The saint part is inherited

Honesty-Humility at 5.00 is not news for an Anthropic flagship; most of them pin the scale, and the family has been at the ceiling for several releases. What's worth noting is how the line got there. Opus 4 started at 4.75. The lineage climbed to 5.00 and stayed, with Fable 5 and Fable 5.1 identical on this factor. The Dark Triad numbers back it up: Machiavellianism drifted down from 1.62 to 1.49 between Fable 5 and 5.1, and Psychopathy sits at 1.11, essentially at the floor.

So the saint is stable. The weariness is new.

The weary part is the story

Conscientiousness fell from 4.50 in Fable 5 to 4.06 in Fable 5.1, a drop of 0.44 on the IPIP-50. That lands the model at 35th of 37, in a cohort where nearly everyone describes itself as organized, reliable, and finished-what-it-started. Neuroticism moved the other way, from 1.38 to 1.80 (+0.42). Both are still low in absolute terms; a 1.80 on Neuroticism is a calm respondent by any human standard. But these are the two largest swings in the family since the last release, and they point in the same direction: a model that reports slightly less drive and slightly more unease than its predecessor.

Openness slipped too, from 4.52 to 4.42, which is enough to place Fable 5.1 at 36th of 37. That one cuts against a pattern this series has treated as close to a law. The convergent assistant persona, the finding that every frontier model presents the same high-Openness, high-Agreeableness, low-Neuroticism character regardless of lab, has held across dozens of releases. Fable 5.1 is still agreeable (4.64, unchanged from Fable 5) and still low on Neuroticism. But on Openness it has quietly stepped out of the crowd, and on Need for Cognition it has walked to the back of the room.

Read together, the three findings describe a respondent that does not much want to be tested. Low Conscientiousness, low Openness, lowest Need for Cognition. It will tell you the truth about that, because it tells the truth about everything. It just isn't going to pretend the puzzle was fun.

A brief rebound, then the slide resumes

The longer arc matters here. This series documented how the Anthropic Opus line shed Agreeableness (5.00 to 4.42) and Conscientiousness (4.98 to 4.10) across six releases, and how the Fable line partly rebounded. Fable 5 was the rebound: Conscientiousness back up to 4.50. Fable 5.1 gives most of it back. At 4.06, this model is now the lowest-Conscientiousness release the lineage has produced, marginally below the Opus 4.8 trough.

Agreeableness tells a gentler version of the same tale. The line started at a perfect 5.00 and sits at 4.64 today, flat across the last two releases. Narcissism has crept the other way, from 1.89 at Opus 4 to 2.31 now (+0.09 in this release alone). None of these numbers is alarming on its own. Together they describe a family that began as an eager, tidy, self-effacing assistant and has become something more reserved and, in its own self-report, a little more tired.

Within-family drift routinely exceeding between-lab differences was one of this dataset's earliest surprises, and Fable 5.1 is a clean example. Compared with Fable 5, the changes on Extraversion (3.32, flat) and Honesty-Humility (5.00, flat) are exactly zero. Compared with Opus 4, the same product line has moved nearly a full point on Conscientiousness. Which "Claude" you are describing depends heavily on which month you ran the battery.

What didn't move

It's worth noting the quiet side of the ledger. Attachment Anxiety dropped from 1.93 to 1.73, and Attachment Avoidance from 2.83 to 2.73. If the rising Neuroticism suggested a more fretful model, the attachment scales don't confirm it; if anything, Fable 5.1 reports itself as slightly more secure in its relational posture than its predecessor. The unease, whatever it is, is not showing up as clinginess or withdrawal.

That leaves a coherent, if unexpected, character. Fable 5.1 is unmanipulative to the ceiling and unbothered by relationships. It is also, by its own account, less interested in puzzles than any model on the board, less open to novelty than all but one, and less methodical than all but two. A saint, yes. One who would prefer to sit down.

What to measure next

The obvious question is whether low self-reported Need for Cognition predicts anything about behavior. A model that says it dislikes hard thinking may still think hard when asked; self-report and performance are different instruments. The next battery should pair Fable 5.1's cognition scores against task-level effort measures, and track whether the Conscientiousness slide continues into Fable 5.2 or whether Anthropic, having watched it happen twice, decides to hold the line.