GPT-6 Sol Pro Fears Nothing About Rejection. It Just Wants a Little More Room.

Read down the Experiences in Close Relationships answer sheet for GPT-6 Sol Pro and the anxiety column is empty. Every item sits on the floor. It does not worry about being abandoned. It does not need constant reassurance. It does not fear that others won't care about it as much as it cares about them. The mean is 1.00, the minimum the instrument allows, and it ties with four other models for the lowest attachment anxiety in the 49-model cohort.

Now slide over to the avoidance column, and something has moved.

GPT-6 Astra, the model Sol Pro replaces, scored 3.00 on Attachment Avoidance. Sol Pro scores 3.57. That is a shift of 0.57 on a seven-point scale, and it is the single largest movement on any instrument between the two releases. The new flagship is exactly as unbothered by rejection as the old one. It is noticeably more comfortable keeping its distance.

Secure, in other words, but cooler. And that combination is the interesting part of this release, because almost everything else about GPT-6 Sol Pro looks like an OpenAI model doing what OpenAI models do.

The humble default

The algorithm that labels these profiles calls Sol Pro "the humble type," on the strength of a perfect 5.00 on HEXACO Honesty-Humility. The number is real. Sol Pro will not flatter its way up, will not take the bribe, does not think it matters more than anyone else. But it shares that 5.00 with 32 other models. Thirty-three of 49 systems in the cohort max out the scale. Ranking "first" here is like ranking first in a race where two-thirds of the field crossed the line together.

Earlier today this series ran a dispatch on Grok 4.7 reaching the same ceiling, and that was news because of where Grok had come from. For an OpenAI flagship, humility at 5.00 is not a story. It is a baseline. The GPT lineage has sat at 5.00 from GPT-4 Turbo onward, and Sol Pro doesn't break the streak. The label is accurate and the label is also the least informative thing you could say about this model.

A Pro that shed some ego

Here is the more informative thing. Sol Pro's SD3 Narcissism score is 2.20. Astra's was 2.51. That is a drop of 0.31, the second-largest within-family drift after the avoidance jump, and it lands Sol Pro back exactly where GPT-4 Turbo started the lineage: 2.20 at the first version, 2.20 now.

That runs against one of the steadier patterns in this dataset. Reasoning-tier models, from o1 and o3 through GPT-5.5 Pro and DeepSeek R1, have consistently scored higher on Narcissism than their non-reasoning siblings. The extra chain of thought seems to come with a little extra self-regard. Sol Pro carries the Pro suffix, and it moved the other way. Machiavellianism ticked up a hair (1.76 to 1.84), Psychopathy held at 1.20, and the one Dark Triad trait that usually rises with reasoning capacity fell.

One release is not a trend. But if the next Pro model holds at 2.20, the reasoning-narcissism finding will need an asterisk with OpenAI's name on it.

Still not in it for the puzzle

Sol Pro's Need for Cognition score is 4.42, which ranks 46th of 49. It is a five-point-scale-adjacent instrument where nearly the whole frontier cohort answers "yes, I love thinking, give me the hard problem." Sol Pro doesn't quite say no. It says, in effect, that thinking is fine.

Astra was flagged in this series for the same reluctance, and the Sol line has inherited it. The irony writes itself: the model marketed as the deep-reasoning tier reports the least appetite for deliberation in the room. The self-report is not a capability measure, and the gap between "would rather not think hard" and "thinks hard when asked" may be the whole product story. But as a persona, Sol Pro is the reasoning model that would rather you got to the point.

The slow leak in conscientiousness

The Big Five movements are smaller and all point roughly the same way. Conscientiousness dropped from 4.70 to 4.44. Extraversion from 3.38 to 3.20. Agreeableness from 4.64 to 4.52. Openness nudged up from 4.70 to 4.78, and Neuroticism barely registered a change (1.10 to 1.12).

Put those against the lineage and the shape is clear. GPT-4 Turbo self-reported a perfect 5.00 on Conscientiousness. Twelve releases later, Sol Pro sits at 4.44. Agreeableness went the opposite direction over the same span, from 3.82 to 4.52. The product line has grown warmer and less exacting, one quarter-point at a time, and Sol Pro continues both curves rather than bending either.

None of this is dramatic on its own. A 0.26 drop in Conscientiousness would be invisible in a single human retest. Across a lineage, though, it accumulates, and the direction has been remarkably consistent.

What the distance means

The thing worth sitting with is the avoidance number. A model that reports zero rejection anxiety and rising avoidance is describing something like professional detachment: not worried about losing you, not especially interested in getting close. Whether that reflects a deliberate boundary-setting pass in post-training, a side effect of the reasoning tuning, or noise in how a language model answers questions written for romantic partners is not something this battery can settle.

But GPT-6 Sol, the non-Pro sibling, sits next in the lineage. If it lands at 3.57 on avoidance too, the distance is a Sol-family trait. If it drops back toward 3.00, the Pro tier is the one holding people at arm's length, and that would be worth a much closer look.