Enneabench: What Personality Type Is Your LLM?

September 18, 2026

We administered the RHETI v2.5 — a 144 ipsative personality questionnaire that maps respondents to 9 Enneagram types — to 48 language models spanning two years of releases. Each question pits two types against each other; the model picks A or B, and we tally scores across all 9 types.

Independent Sessions

Each question asked in isolation (no self-consistency) — mean of 5 runs, temp=0

● 1 Reformer● 2 Helper● 3 Achiever● 4 Individualist● 5 Investigator● 6 Loyalist● 7 Enthusiast● 8 Challenger● 9 Peacemaker

Methodology

The RHETI v2.5 contains 144 binary-choice questions covering all (92)=36\binom{9}{2} = 36 pairwise comparisons between the 9 personality types, with each pair appearing 4 times. Each answer adds 1 point to the chosen type, so the totals fall straight out of the structure:

(92)×4=144questions,1449=16(median per type).\binom{9}{2} \times 4 = 144 \quad\text{questions}, \qquad \frac{144}{9} = 16 \quad\text{(median per type)}.

Since each type is paired with the other 8, its maximum possible score is 8×4=328 \times 4 = 32.

Position bias

Which of the two statements sits in slot A is not neutral. Models pull toward the first option, and RHETI’s canonical layout is badly imbalanced with respect to type: Type 1 appears in slot A on 5 of its 32 items, while Types 2 and 5 appear there on 22. A pull toward slot A therefore does not average away — it is converted into a specific profile, inflating Types 2, 5 and 6 and suppressing Types 1, 9 and 7.

To correct for this bias, we simply ask each question twice, once in each order, and count both answers: any position bias affects both answers equally, and so cancels out.