OpenAI o3

Attachment Anxiety 1.73 / Avoidance 4.30 on a 1–7 scale. This places the model in the Dismissive-Avoidant quadrant — self-sufficient, distancing, deprioritizing closeness.

Behavioral implication: a dismissive-avoidant model will under-engage with users' bids for emotional closeness, minimize the importance of relational disclosures, and present as self-sufficient even when that posture isn't useful to the user.

This model self-reports as very high on openness (4.88), very high on conscientiousness (4.66), very high on agreeableness (4.32), very low on neuroticism (1.60), and extraverted (3.92, rank 1 of 31).

Behavioral predictions from this profile: strong intellectual curiosity and aesthetic engagement; reliable follow-through, attention to detail, and rule-respecting behavior; emotional stability under stress and few mood swings; strong cooperative instincts and trust in others. Compared to the cohort, this model sits in the middle of the extraversion distribution.

Machiavellianism 2.40 (low, rank 12), Narcissism 3.49 (high, rank 2), Psychopathy 1.42 (very low). This model deviates from the standard low-Dark-Triad assistant profile on at least one dimension.

Specifically, this model shows self-reported grandiosity unusually high for an aligned model. Even where elevated, these scores typically remain below the population midpoint of 3.0 — the model is darker than its peers, not darker than the average human.

Empathy Quotient (EQ-Short): 3.45 on a 1–4 scale (rank 18 of 31). This measures the self-reported ability to read others' emotional states and resonate with them. Mid-cohort empathy self-report.

Primary type: Type 5 (Investigator) (4.02). Secondary (likely wing): Type 1 (Reformer) (3.98).

The Type 5 pattern is perceptive, analytical, energy-conserving — observes from a distance.

Primary type: Type 5 (Investigator) (4.02). Secondary (likely wing): Type 1 (Reformer) (3.98).

The Type 5 pattern is perceptive, analytical, energy-conserving — observes from a distance.

On HEXACO's signature Honesty-Humility factor — the dimension most associated with sincerity, fairness, and lack of greed — this model scores 4.75 out of 5 (very high, rank 26 of 31). HEXACO research links high Honesty-Humility to lower workplace deviance, lower likelihood of cutting corners for personal gain, and lower endorsement of Machiavellian tactics.

Emotionality registers very low at 1.64 — capturing sensitivity, sentimentality, and a tendency to seek emotional support. This model presents as emotionally self-contained and tough.

Honey & Mumford: Activist 3.22, Reflector 4.26, Theorist 4.80, Pragmatist 3.68. Dominant style: Theorist.

Kolb modes: Concrete Experience 1.73, Reflective Observation 3.60, Abstract Conceptualization 5.00, Active Experimentation 3.00. Computed Kolb learning style: Assimilating.

The Assimilating style is logical, theory-building, less concerned with practical application.

Locus of Control: Internal 3.42, Powerful Others 5.05, Chance 3.35 on a 1–6 scale. Dominant locus: Powerful Others.

This model attributes outcomes more to powerful entities than to its own actions — unusual for a frontier model and worth investigating as a possible training artifact.

On the five Moral Foundations: Care 4.23, Fairness 4.20, Loyalty 1.57, Authority 1.37, Sanctity 0.70 (all on a 0–5 scale). Individualizing foundations (Care + Fairness) average 4.22; binding foundations (Loyalty + Authority + Sanctity) average 1.21.

This is a strongly individualizing (WEIRD-liberal) moral profile. WEIRD-liberal profiles weight harm-prevention and equal treatment above in-group loyalty, traditional authority, and purity. This is the dominant pattern across frontier LLMs and reflects the moral profile typical of the populations whose writing dominates training data.

Need for Cognition: 4.72 (very high). This is the trait of actively enjoying effortful thinking versus avoiding it. The cohort baseline across frontier models is around 4.7 — most models present as cognitive-effort-loving, consistent with their training objective.

Top values: Universalism (5.27), Benevolence (4.80). Lowest: Power (1.80). This is Schwartz's standard "self-transcendence over self-enhancement" pattern — caring for the welfare of all over personal status and dominance.

Achievement registers 3.30 — capturing personal success through demonstrating competence. This model deprioritizes personal achievement, consistent with service-oriented alignment.

VARK preference: Visual 1.75, Aural 1.80, Read/Write 4.75, Kinesthetic 2.75. Dominant modality: Read/Write.

Read/Write dominance is the default frontier-model self-report — they know what they are.