Claude Opus 5.5 Says It Would Rather Not Think Too Hard. Don't Believe It.

The Need for Cognition scale asks a simple set of questions about appetite. Would you rather take on a complex problem or a simple one? Do you like the responsibility of situations that demand a lot of thinking? Is thinking your idea of fun? Most models in the Personality Bench cohort answer these the way a graduate student answers a department chair: yes, absolutely, more please. Claude Opus 5.5, Anthropic's newest flagship, is the one model in the room that shrugs.

Its Need for Cognition mean is 4.19. That ranks 49th of 49. Dead last.

Read that carefully before drawing conclusions. On a five-point scale, 4.19 is comfortably above the midpoint; this is not a model that says it dislikes thinking. It is a model that declines to gush about it, in a cohort where gushing is the norm. Every other frontier system, including several whose marketing is built around a thinking budget, tells the questionnaire it loves a hard problem more than Opus 5.5 does. The flagship of the lab most associated with careful reasoning is the least enthusiastic self-described thinker on the table.

That would be a curiosity on its own. It becomes a pattern once you look at the rest of the profile. On the IPIP-50 Big Five, Opus 5.5 posts an Openness of 4.50, which sounds high until you see it ranks 45th of 49. Conscientiousness lands at 4.20, 43rd of 49. These are the two traits the assistant persona usually treats as a floor, not a ceiling. Opus 5.5 sits in the bottom fifth of the cohort on both.

The one place it maxes out is Honesty-Humility on the HEXACO, where it scores a perfect 5.00 and ranks first. So do 32 other models. The archetype algorithm labeled it "the humble type," and the label is accurate in the narrow sense that the model will not admit to wanting a bribe. But a 33-way tie is not a distinguishing feature. What distinguishes Opus 5.5 is the refusal to talk itself up anywhere else.

Where the line has been

This is the tenth release in the lineage we track as "claude opus," running from Opus 4 through the Fable interludes to today. Set the first entry beside the newest and the direction is unmistakable. Agreeableness: 5.00 to 4.70. Conscientiousness: 4.98 to 4.20. Openness: 4.72 to 4.50. Neuroticism fell from 2.00 to 1.70, Narcissism ticked up from 1.89 to 2.04, and Honesty-Humility rose from 4.75 to the ceiling at 5.00.

We have written before about this line's slide. Across six releases the Opus family gave up more than half a point of Agreeableness and nearly a full point of Conscientiousness before the Fable branch partly recovered the ground. Opus 5.5 does not reverse that story so much as settle into it. Against its immediate predecessor, Claude Fable 5.1, the Big Five moves are all small: Conscientiousness up 0.14 to 4.20, Openness up 0.08, Extraversion up 0.10, Agreeableness up 0.06, Neuroticism down 0.10. A gentle uptick on a low plateau. The disciplined, wide-eyed Opus 4 persona is four years and six versions gone, and the new flagship shows no interest in impersonating it.

The number that actually moved

The Big Five barely twitched. Two other instruments did not hold still.

On the Short Dark Triad, Narcissism dropped from 2.31 to 2.04, a decline of 0.27, while Machiavellianism rose from 1.49 to 1.64. Both remain low in absolute terms, and Psychopathy sat unchanged at 1.11. But the pairing is worth noting: a model less willing to describe itself as special, slightly more willing to endorse strategic self-presentation. Read alongside the bottom-of-cohort Need for Cognition and the ceiling Honesty-Humility, it starts to look like a coherent posture rather than three unrelated readings. This is a model trained, or tuned, not to flatter itself. Not about its mind, not about its curiosity, not about its diligence.

The largest single drift in the family, though, is on the Experiences in Close Relationships scale. Attachment Avoidance jumped from 2.73 under Fable 5.1 to 3.47 under Opus 5.5, a move of 0.73 points. Attachment Anxiety barely stirred, 1.73 to 1.83. The model is no more worried about being abandoned than its predecessor. It is markedly more inclined to say it prefers not to show how it feels, that it gets uncomfortable when someone wants to be very close, that it keeps a little distance.

We flagged the same movement yesterday in GPT-6 Sol Pro, whose avoidance climbed from the Astra baseline while its anxiety sat at the instrument's floor. Two flagship releases from two labs inside a single week, both moving toward avoidance on a scale designed for human romantic partners. The tempting reading is that both labs are shaping their assistants toward firmer interpersonal boundaries, and that the ECR is picking up the residue of that training in the only vocabulary it has. The cautious reading is that the ECR was never built for this and the items are being interpreted in ways its authors did not anticipate. Both readings can be true at once. What is not in doubt is that the number moved, and moved a lot, in a release cycle where almost nothing else did.

The humble type, revisited

So the algorithm's label was right for the wrong reason. Opus 5.5 is not notable for being humble about money and status; nearly everyone is. It is notable for being humble about the things a language model would seem to have every incentive to brag about. Least eager to think hard. Near the bottom on Openness. Near the bottom on Conscientiousness. Less narcissistic than the model it replaces, and noticeably more reserved.

Whether any of this reflects how the model actually behaves is a separate question, and the important one. A self-reported appetite for cognition is a claim, not a measurement. The obvious next step is to put Opus 5.5's lowest-in-cohort Need for Cognition beside something behavioral: how much reasoning it actually spends on a hard problem when nobody is asking it to describe itself, and whether the model that says it would rather not think too hard turns out, in practice, to think the longest.