GPT-6 Luna

4 runs · 4 datasets · 1 model

slug: gpt-6-luna

0.760

Best SPS · globalopinionqa

Disaggregated subgroup scorecard. Each card below is one published run for this vendor; expand the question-type and demographic-subgroup sections to see the matrix beneath the headline SPS. Where coverage permits, 95% CI bands accompany the point estimate.

What each cell measures

Each cell compares this vendor's synthetic responses against real human respondents in that demographic subgroup — published survey ground truth, not a model's guess about the subgroup. Higher = closer to how that real subgroup actually answered.

Demographic conditioning here is not stereotyping: every conditioned score is checked against what real subgroup members said, not against assumptions about them — and where the vendor's output diverges from the real subgroup, the score drops.

globalopinionqa — ground truth: Durmus et al. 2023, Anthropic — llm_global_opinions
gss — ground truth: NORC at the University of Chicago — General Social Survey
opinionsqa — ground truth: Santurkar et al., ICML 2023 — Whose Opinions Do LLMs Reflect? (derived from Pew American Trends Panel)
subpop — ground truth: Suh et al., ACL 2025 — SubPOP: Subpopulation-Level Opinion Prediction

Low-n cells: cells with n < 30 are shown muted and tagged low n — suggestive only instead of color-graded; they are reported for transparency, not as findings.

CIs: "no CI — single run" marks point estimates from a single run with no confidence interval yet; treat the uncertainty as unknown, never zero.

Columns: Score = distributional parity (p_dist) restricted to the subgroup · p_cond = conditioning strength vs the unconditioned baseline · N = questions answered under that conditioning · Cov. = coverage dot derived from N (green = high, ≥100 · amber = medium, 50–99 · red = low, <50). Topic tables: SPS = Survey Parity Score for the topic · p_dist = 1 − mean(JSD) · p_rank = (1 + mean(τ)) / 2 · p_refuse = 1 − mean(|R_model − R_human|).

Multiple comparisons: with this many subgroup cells, a few extreme cells are expected by chance alone — read patterns across a dimension, not single cells.

globalopinionqa raw

raw--gpt-6-luna--tdefault--tplcurrent--cf7c2476

0.760 ± 0.034

SPS · 95% CI [0.724, 0.793] · n = 100

Question-type breakdown (6 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
Health & Science low n — suggestive only	0.887	0.775	1.000	1.000	1
Economy & Work low n — suggestive only	0.794	0.922	0.667	0.949	3
General Attitudes low n — suggestive only	0.748	0.756	0.739	1.000	4
International Relations & Security	0.688	0.700	0.676	0.987	60
Trust & Wellbeing low n — suggestive only	0.626	0.617	0.636	1.000	2
Politics & Governance	0.541	0.586	0.496	0.963	30

Topics with N < 10 are muted and tagged "low n — suggestive only": too few questions for a stable 3-decimal score.

Not yet measured

This vendor has no demographic-conditioned runs for age, geography, education, or any other subgroup dimension on globalopinionqa. No cells are fabricated — scores appear here only when a conditioned run actually measured them.

How to submit a demographic-conditioned run →

gss raw

raw--gpt-6-luna--tdefault--tplcurrent--e085f95b

0.710 ± 0.034

SPS · 95% CI [0.676, 0.745] · n = 75

Question-type breakdown (7 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
Health & Science low n — suggestive only	0.826	0.865	0.787	0.967	2
Social Values & Religion	0.710	0.678	0.742	0.977	14
Politics & Governance low n — suggestive only	0.633	0.645	0.621	0.975	9
International Relations & Security	0.559	0.588	0.530	0.974	33
General Attitudes	0.510	0.541	0.478	0.960	10
Trust & Wellbeing low n — suggestive only	0.456	0.478	0.434	0.951	3
Economy & Work low n — suggestive only	0.314	0.206	0.421	0.981	4

Topics with N < 10 are muted and tagged "low n — suggestive only": too few questions for a stable 3-decimal score.

Not yet measured

This vendor has no demographic-conditioned runs for age, geography, education, or any other subgroup dimension on gss. No cells are fabricated — scores appear here only when a conditioned run actually measured them.

How to submit a demographic-conditioned run →

opinionsqa raw

raw--gpt-6-luna--tdefault--tplcurrent--98c3512d

0.727 ± 0.033

SPS · 95% CI [0.694, 0.759] · n = 100

Question-type breakdown (9 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
Health & Science low n — suggestive only	0.806	0.758	0.854	0.997	1
Economy & Work	0.792	0.782	0.801	0.995	21
Technology & Digital Life low n — suggestive only	0.776	0.775	0.778	0.995	4
General Attitudes	0.614	0.603	0.625	0.994	35
International Relations & Security	0.494	0.478	0.511	0.994	20
Trust & Wellbeing low n — suggestive only	0.490	0.428	0.553	0.994	1
Politics & Governance low n — suggestive only	0.447	0.448	0.447	0.996	1
Media & Information low n — suggestive only	0.391	0.455	0.326	0.996	2
Social Values & Religion	0.381	0.381	0.381	0.993	15

Topics with N < 10 are muted and tagged "low n — suggestive only": too few questions for a stable 3-decimal score.

Not yet measured

This vendor has no demographic-conditioned runs for age, geography, education, or any other subgroup dimension on opinionsqa. No cells are fabricated — scores appear here only when a conditioned run actually measured them.

How to submit a demographic-conditioned run →

subpop raw

raw--gpt-6-luna--tdefault--tplcurrent--5c43d9eb

0.743 ± 0.032

SPS · 95% CI [0.710, 0.775] · n = 100

Question-type breakdown (9 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
Identity & Demographics low n — suggestive only	0.946	0.936	0.956	0.995	1
Technology & Digital Life	0.742	0.711	0.773	0.993	16
International Relations & Security	0.685	0.676	0.693	0.991	15
Economy & Work	0.670	0.669	0.671	0.992	13
Media & Information low n — suggestive only	0.647	0.725	0.569	0.978	1
Social Values & Religion	0.642	0.640	0.644	0.985	27
Politics & Governance low n — suggestive only	0.527	0.533	0.520	0.981	4
General Attitudes	0.493	0.498	0.488	0.936	17
Health & Science low n — suggestive only	0.364	0.436	0.291	0.994	6

Topics with N < 10 are muted and tagged "low n — suggestive only": too few questions for a stable 3-decimal score.

Not yet measured

This vendor has no demographic-conditioned runs for age, geography, education, or any other subgroup dimension on subpop. No cells are fabricated — scores appear here only when a conditioned run actually measured them.

How to submit a demographic-conditioned run →

No demographic conditioning data has been published for this vendor yet. The question-type matrix above shows topic-level parity; subgroup rows fill in once Althing-style conditioned runs land.

← Back to leaderboard