Ensemble: Qwen 3.8 Flash, DeepSeek V4.1 Flash, Gemini 3.5 Flash-Lite (Althing)

4 runs · 4 datasets · 1 model

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

0.887

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--flash3-althing--tdefault--tplcurrent--6c4d3cb1

0.811 ± 0.029

SPS · 95% CI [0.781, 0.839] · n = 100

Question-type breakdown (6 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
Health & Science low n — suggestive only	1.000	1.000	1.000	1.000	1
Economy & Work low n — suggestive only	0.936	0.903	0.969	0.949	3
Trust & Wellbeing low n — suggestive only	0.785	0.699	0.871	1.000	2
General Attitudes low n — suggestive only	0.774	0.809	0.739	1.000	4
International Relations & Security	0.719	0.742	0.696	0.987	60
Politics & Governance	0.702	0.712	0.692	0.958	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--flash3-althing--tdefault--tplcurrent--315646e2

0.848 ± 0.022

SPS · 95% CI [0.826, 0.871] · n = 75

Question-type breakdown (7 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
Economy & Work low n — suggestive only	0.857	0.799	0.915	0.981	4
Health & Science low n — suggestive only	0.829	0.870	0.787	0.967	2
Social Values & Religion	0.814	0.800	0.829	0.977	14
International Relations & Security	0.801	0.791	0.810	0.974	33
Politics & Governance low n — suggestive only	0.782	0.761	0.803	0.975	9
General Attitudes	0.726	0.713	0.740	0.960	10
Trust & Wellbeing low n — suggestive only	0.595	0.590	0.600	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--flash3-althing--tdefault--tplcurrent--ed260b81

0.887 ± 0.014

SPS · 95% CI [0.872, 0.900] · n = 100

Question-type breakdown (9 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
Media & Information low n — suggestive only	0.971	0.983	0.959	0.996	2
Social Values & Religion	0.886	0.878	0.894	0.993	15
Health & Science low n — suggestive only	0.881	0.805	0.956	0.997	1
Technology & Digital Life low n — suggestive only	0.866	0.840	0.892	0.995	4
Economy & Work	0.850	0.836	0.863	0.995	21
International Relations & Security	0.817	0.827	0.806	0.994	20
Politics & Governance low n — suggestive only	0.802	0.806	0.799	0.996	1
General Attitudes	0.800	0.785	0.815	0.994	35
Trust & Wellbeing low n — suggestive only	0.743	0.669	0.816	0.994	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--flash3-althing--tdefault--tplcurrent--82016234

0.847 ± 0.022

SPS · 95% CI [0.824, 0.868] · n = 100

Question-type breakdown (9 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
Health & Science low n — suggestive only	0.883	0.902	0.864	0.994	6
Identity & Demographics low n — suggestive only	0.872	0.910	0.833	0.995	1
General Attitudes	0.811	0.801	0.821	0.936	17
Economy & Work	0.810	0.776	0.845	0.992	13
Politics & Governance low n — suggestive only	0.810	0.803	0.816	0.984	4
Technology & Digital Life	0.768	0.738	0.798	0.993	16
International Relations & Security	0.766	0.754	0.777	0.991	15
Social Values & Religion	0.737	0.709	0.765	0.985	27
Media & Information low n — suggestive only	0.574	0.518	0.629	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