Claude Haiku 5.5

4 runs · 4 datasets · 1 model

slug: claude-haiku-5-5

0.812

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 raw

raw--claude-haiku-5-5--tdefault--tplcurrent--b1dcdac0

0.701 ± 0.037

SPS · 95% CI [0.664, 0.737] · 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	0.794	0.921	0.667	0.949	3
General Attitudes low n — suggestive only	0.788	0.838	0.739	0.975	4
Trust & Wellbeing low n — suggestive only	0.744	0.668	0.819	0.942	2
International Relations & Security	0.640	0.660	0.620	0.914	60
Politics & Governance	0.555	0.594	0.517	0.659	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--claude-haiku-5-5--tdefault--tplcurrent--a7a35ba7

0.695 ± 0.042

SPS · 95% CI [0.653, 0.737] · n = 75

Question-type breakdown (7 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
International Relations & Security	0.618	0.646	0.589	0.974	33
Social Values & Religion	0.575	0.568	0.583	0.970	14
Politics & Governance low n — suggestive only	0.547	0.549	0.545	0.941	9
Health & Science low n — suggestive only	0.517	0.533	0.500	0.967	2
Trust & Wellbeing low n — suggestive only	0.482	0.531	0.434	0.951	3
General Attitudes	0.441	0.455	0.428	0.960	10
Economy & Work low n — suggestive only	0.430	0.398	0.461	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--claude-haiku-5-5--tdefault--tplcurrent--d27ebf59

0.812 ± 0.023

SPS · 95% CI [0.789, 0.835] · n = 100

Question-type breakdown (9 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
Media & Information low n — suggestive only	0.902	0.897	0.908	0.996	2
Health & Science low n — suggestive only	0.820	0.866	0.774	0.997	1
Technology & Digital Life low n — suggestive only	0.786	0.756	0.816	0.995	4
International Relations & Security	0.786	0.783	0.788	0.994	20
Economy & Work	0.758	0.745	0.771	0.995	21
General Attitudes	0.698	0.678	0.719	0.993	35
Social Values & Religion	0.650	0.643	0.658	0.993	15
Politics & Governance low n — suggestive only	0.446	0.591	0.300	0.996	1
Trust & Wellbeing low n — suggestive only	0.106	0.028	0.184	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 raw

raw--claude-haiku-5-5--tdefault--tplcurrent--aee6d676

0.726 ± 0.032

SPS · 95% CI [0.692, 0.756] · n = 100

Question-type breakdown (9 topics)

Topic	SPS	p_dist	p_rank	p_refuse	N
Economy & Work	0.709	0.696	0.722	0.992	13
Technology & Digital Life	0.704	0.671	0.737	0.993	16
Identity & Demographics low n — suggestive only	0.701	0.811	0.591	0.995	1
General Attitudes	0.695	0.717	0.672	0.934	17
International Relations & Security	0.644	0.614	0.674	0.991	15
Politics & Governance low n — suggestive only	0.525	0.503	0.548	0.962	4
Social Values & Religion	0.459	0.435	0.484	0.957	27
Media & Information low n — suggestive only	0.457	0.458	0.457	0.978	1
Health & Science low n — suggestive only	0.457	0.465	0.449	0.996	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