Gemini Flash Lite

3 runs · 3 datasets · 1 model

slug: gemini-flash-lite

0.776

Best SPS · subpop

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
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--gemini-2.5-flash-lite--tdefault--tplcurrent--ddafa278

0.761 ± 0.032

SPS · 95% CI [0.729, 0.792] · 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	0.730	0.793	0.667	0.949	3
International Relations & Security	0.665	0.689	0.641	0.986	60
Politics & Governance	0.631	0.659	0.602	0.962	30
General Attitudes low n — suggestive only	0.594	0.789	0.398	1.000	4
Trust & Wellbeing low n — suggestive only	0.490	0.457	0.523	1.000	2

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 →

opinionsqa raw

raw--gemini-2.5-flash-lite--tdefault--tplcurrent--bdb40b06

0.738 ± 0.032

SPS · 95% CI [0.704, 0.769] · n = 100

Question-type breakdown (8 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
Social Values & Religion	0.799	0.794	0.804	—	15
Health & Science low n — suggestive only	0.768	0.763	0.774	—	1
Economy & Work	0.765	0.750	0.781	—	20
Technology & Digital Life low n — suggestive only	0.764	0.736	0.792	—	4
Politics & Governance low n — suggestive only	0.763	0.728	0.799	—	1
General Attitudes	0.736	0.739	0.733	—	38
Media & Information low n — suggestive only	0.701	0.744	0.658	—	2
International Relations & Security	0.660	0.661	0.659	—	19

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--gemini-2.5-flash-lite--tdefault--tplcurrent--46e46626

0.776 ± 0.029

SPS · 95% CI [0.746, 0.803] · n = 100

Question-type breakdown (9 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
Identity & Demographics low n — suggestive only	0.846	0.838	0.854	0.995	1
Technology & Digital Life	0.735	0.700	0.771	0.993	16
International Relations & Security	0.730	0.712	0.747	0.991	15
Health & Science low n — suggestive only	0.693	0.641	0.744	0.994	6
General Attitudes	0.682	0.693	0.671	0.932	17
Economy & Work	0.678	0.629	0.727	0.992	13
Social Values & Religion	0.605	0.587	0.624	0.985	27
Media & Information low n — suggestive only	0.604	0.579	0.629	0.978	1
Politics & Governance low n — suggestive only	0.602	0.585	0.620	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 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