Ensemble: Qwen 3.8 Flash (Althing), Sonnet 4 (Althing), DeepSeek V4.1 Flash, Gemini 2.5 Flash-Lite

4 runs · 4 datasets · 1 model

slug: ensemble-qwen-3-8-flash-althing-sonnet-4-althing-deepseek-v4-1-flash-gemini-2-5-flash-lite

0.913

Best SPS · opinionsqa

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 product

ensemble--complementary-4--tdefault--tplcurrent--2e4cb430

0.831 ± 0.027

SPS · 95% CI [0.804, 0.858] · n = 100

Question-type breakdown (6 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
Health & Science low n — suggestive only	0.976	0.952	1.000	1.000	1
Economy & Work low n — suggestive only	0.940	0.910	0.969	0.949	3
General Attitudes low n — suggestive only	0.804	0.900	0.708	1.000	4
Trust & Wellbeing low n — suggestive only	0.787	0.746	0.828	0.996	2
International Relations & Security	0.765	0.798	0.732	0.987	60
Politics & Governance	0.705	0.761	0.649	0.962	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 product

ensemble--complementary-4--tdefault--tplcurrent--fb17ebf4

0.863 ± 0.022

SPS · 95% CI [0.841, 0.885] · n = 75

Question-type breakdown (7 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
International Relations & Security	0.837	0.845	0.828	0.974	33
Health & Science low n — suggestive only	0.835	0.882	0.787	0.967	2
Social Values & Religion	0.808	0.813	0.804	0.977	14
Politics & Governance low n — suggestive only	0.791	0.780	0.801	0.975	9
General Attitudes	0.779	0.798	0.761	0.960	10
Economy & Work low n — suggestive only	0.759	0.765	0.753	0.981	4
Trust & Wellbeing low n — suggestive only	0.700	0.732	0.668	0.951	3

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 product

ensemble--complementary-4--tdefault--tplcurrent--1c6ab6e1

0.913 ± 0.013

SPS · 95% CI [0.900, 0.926] · n = 100

Question-type breakdown (9 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
Media & Information low n — suggestive only	0.968	0.978	0.959	0.996	2
Health & Science low n — suggestive only	0.927	0.897	0.956	0.997	1
Trust & Wellbeing low n — suggestive only	0.915	0.912	0.918	0.994	1
Economy & Work	0.900	0.884	0.915	0.995	21
Social Values & Religion	0.899	0.901	0.897	0.993	15
Technology & Digital Life low n — suggestive only	0.875	0.862	0.888	0.995	4
General Attitudes	0.855	0.860	0.849	0.994	35
International Relations & Security	0.844	0.865	0.823	0.994	20
Politics & Governance low n — suggestive only	0.760	0.721	0.799	0.996	1

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 product

ensemble--complementary-4--tdefault--tplcurrent--d17dc795

0.865 ± 0.022

SPS · 95% CI [0.841, 0.885] · n = 100

Question-type breakdown (9 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
Health & Science low n — suggestive only	0.902	0.939	0.864	0.994	6
Identity & Demographics low n — suggestive only	0.890	0.947	0.833	0.995	1
Economy & Work	0.855	0.825	0.886	0.992	13
Politics & Governance low n — suggestive only	0.843	0.864	0.823	0.983	4
General Attitudes	0.828	0.825	0.831	0.936	17
International Relations & Security	0.786	0.800	0.773	0.991	15
Social Values & Religion	0.782	0.773	0.791	0.985	27
Technology & Digital Life	0.773	0.743	0.803	0.993	16
Media & Information low n — suggestive only	0.597	0.620	0.575	0.978	1

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