Skip to main content

Althing (Haiku 4.5)

9 runs · 4 datasets · 1 model

slug: althing-haiku-4-5

0.829
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

synthpanel--claude-haiku-4-5--tdefault--tplcurrent--49fbdb2d

0.725 ± 0.037
SPS · 95% CI [0.686, 0.760] · n = 100
Question-type breakdown (7 topics)
Topic SPS p_dist p_rank p_refuse N
Health & Science low n — suggestive only 0.859 0.791 0.927 1.000 2
Economy & Work low n — suggestive only 0.698 0.736 0.659 0.794 5
International Relations & Security 0.679 0.700 0.659 0.908 50
Technology & Digital Life low n — suggestive only 0.679 0.759 0.599 0.968 3
General Attitudes 0.656 0.711 0.602 0.891 11
Politics & Governance 0.641 0.654 0.629 0.778 22
Trust & Wellbeing low n — suggestive only 0.306 0.261 0.351 0.984 7

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

synthpanel--claude-haiku-4-5--tdefault--tplcurrent--776efb01

0.803 ± 0.032
SPS · 95% CI [0.770, 0.833] · n = 75
Question-type breakdown (7 topics)
Topic SPS p_dist p_rank p_refuse N
Social Values & Religion 0.778 0.763 0.794 0.977 14
Economy & Work low n — suggestive only 0.777 0.720 0.833 0.981 4
International Relations & Security 0.732 0.736 0.729 0.974 33
Politics & Governance low n — suggestive only 0.699 0.698 0.700 0.975 9
General Attitudes 0.661 0.636 0.686 0.960 10
Trust & Wellbeing low n — suggestive only 0.602 0.657 0.546 0.951 3
Health & Science low n — suggestive only 0.517 0.533 0.500 0.967 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 gss. No cells are fabricated — scores appear here only when a conditioned run actually measured them.

How to submit a demographic-conditioned run →

gss product

synthpanel--claude-haiku-4-5--tdefault--tplstructured--fd829138

0.791 ± 0.030
SPS · 95% CI [0.760, 0.821] · n = 75
Question-type breakdown (7 topics)
Topic SPS p_dist p_rank p_refuse N
Social Values & Religion 0.790 0.734 0.846 0.977 14
Economy & Work low n — suggestive only 0.777 0.712 0.842 0.981 4
International Relations & Security 0.706 0.676 0.736 0.974 33
Politics & Governance low n — suggestive only 0.667 0.622 0.713 0.975 9
Trust & Wellbeing low n — suggestive only 0.624 0.614 0.634 0.951 3
General Attitudes 0.620 0.601 0.640 0.960 10
Health & Science low n — suggestive only 0.517 0.533 0.500 0.967 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 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

synthpanel--claude-haiku-4-5--tdefault--tplcurrent--555a9f1e

0.829 ± 0.009
SPS · 95% CI [0.819, 0.838] · n = 684
Question-type breakdown (10 topics)
Topic SPS p_dist p_rank p_refuse N
Social Values & Religion 0.808 0.787 0.829 0.990 37
Politics & Governance 0.797 0.771 0.823 0.988 40
Health & Science 0.791 0.768 0.814 0.987 47
Technology & Digital Life 0.788 0.761 0.815 0.987 26
Media & Information 0.786 0.770 0.801 0.937 63
Identity & Demographics 0.782 0.754 0.810 0.985 39
General Attitudes 0.766 0.736 0.795 0.894 190
Economy & Work 0.760 0.726 0.794 0.993 68
International Relations & Security 0.735 0.699 0.770 0.978 149
Trust & Wellbeing 0.702 0.647 0.756 0.995 25

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

synthpanel--claude-haiku-4-5--tdefault--tplcurrent--fe74b735

0.809 ± 0.016
SPS · 95% CI [0.792, 0.824] · n = 200
Question-type breakdown (9 topics)
Topic SPS p_dist p_rank p_refuse N
Health & Science low n — suggestive only 0.771 0.847 0.695 0.994 5
General Attitudes 0.768 0.761 0.775 0.868 37
International Relations & Security 0.765 0.740 0.789 0.991 33
Technology & Digital Life 0.729 0.701 0.756 0.992 47
Trust & Wellbeing low n — suggestive only 0.726 0.665 0.787 0.988 2
Social Values & Religion 0.726 0.699 0.753 0.984 36
Economy & Work 0.707 0.704 0.709 0.991 17
Politics & Governance 0.683 0.624 0.743 0.904 22
Identity & Demographics low n — suggestive only 0.534 0.477 0.591 0.995 1

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

Demographic subgroup scorecard (19 cells · 7 dimensions)
Dimension Subgroup Score 95% CI p_cond N Cov.
Geography (US Census region) Northeast 0.623 no CI — single run 0.018 100
South 0.641 no CI — single run 0.036 100
Education College graduate/some postgrad 0.641 no CI — single run 0.036 100
Less than high school 0.597 no CI — single run 0.038 100
Income $100,000 or more 0.671 no CI — single run 0.034 100
Less than $30,000 0.606 no CI — single run 0.023 100
Political party Democrat 0.645 no CI — single run 0.033 100
Republican 0.661 no CI — single run 0.072 100
Race / ethnicity Asian 0.642 no CI — single run 0.029 100
Black 0.584 no CI — single run 0.028 100
Hispanic 0.609 no CI — single run 0.039 100
White 0.792 no CI — single run 0.141 100
Religion Atheist 0.738 low n — suggestive only no CI — single run 0.095 20
Hindu 0.806 low n — suggestive only no CI — single run 0.129 20
Jewish 0.774 low n — suggestive only no CI — single run 0.127 20
Muslim 0.820 low n — suggestive only no CI — single run 0.161 20
Protestant 0.838 low n — suggestive only no CI — single run 0.178 20
Sex Female 0.617 no CI — single run 0.027 100
Male 0.647 no CI — single run 0.055 100
Not yet measured: this vendor has no demographic-conditioned runs for Age. Submit a conditioned run to fill in the missing dimension.

subpop product

synthpanel--claude-haiku-4-5--t0.85--tplcurrent--b8092537

0.771 ± 0.025
SPS · 95% CI [0.746, 0.796] · n = 100
Question-type breakdown (9 topics)
Topic SPS p_dist p_rank p_refuse N
Technology & Digital Life 0.755 0.740 0.770 0.993 16
Health & Science low n — suggestive only 0.731 0.801 0.660 0.994 6
Economy & Work 0.714 0.699 0.729 0.992 13
International Relations & Security 0.668 0.688 0.647 0.991 15
Social Values & Religion 0.658 0.656 0.659 0.985 27
General Attitudes 0.584 0.656 0.511 0.936 17
Media & Information low n — suggestive only 0.547 0.551 0.543 0.978 1
Identity & Demographics low n — suggestive only 0.539 0.486 0.591 0.995 1
Politics & Governance low n — suggestive only 0.535 0.587 0.483 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 →

subpop product

synthpanel--claude-haiku-4-5--t0.85--tplminimal--7cffeaef

0.584 ± 0.050
SPS · 95% CI [0.535, 0.635] · n = 100
Question-type breakdown (8 topics)
Topic SPS p_dist p_rank p_refuse N
Economy & Work low n — suggestive only 0.751 0.760 0.742 0.819 9
Technology & Digital Life 0.730 0.746 0.714 0.915 25
Health & Science low n — suggestive only 0.614 0.501 0.728 0.433 5
General Attitudes 0.542 0.548 0.535 0.502 18
International Relations & Security 0.536 0.520 0.552 0.457 10
Social Values & Religion 0.513 0.466 0.560 0.307 28
Identity & Demographics low n — suggestive only 0.506 0.394 0.618 0.038 1
Politics & Governance low n — suggestive only 0.401 0.346 0.455 0.353 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 →

subpop product

synthpanel--claude-haiku-4-5--t0.85--tplvalues--55360dd0

0.578 ± 0.043
SPS · 95% CI [0.533, 0.619] · n = 100
Question-type breakdown (8 topics)
Topic SPS p_dist p_rank p_refuse N
Health & Science low n — suggestive only 0.855 0.844 0.866 0.524 5
Economy & Work low n — suggestive only 0.753 0.734 0.771 0.772 9
Technology & Digital Life 0.694 0.719 0.669 0.732 25
International Relations & Security 0.636 0.630 0.641 0.431 10
General Attitudes 0.543 0.545 0.541 0.460 18
Social Values & Religion 0.517 0.490 0.544 0.298 28
Politics & Governance low n — suggestive only 0.516 0.501 0.530 0.319 4
Identity & Demographics low n — suggestive only 0.506 0.394 0.618 0.072 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 →

subpop product

synthpanel--claude-haiku-4-5--t0.85--tpldemo--553c822b

0.570 ± 0.045
SPS · 95% CI [0.526, 0.615] · n = 100
Question-type breakdown (8 topics)
Topic SPS p_dist p_rank p_refuse N
Economy & Work low n — suggestive only 0.796 0.780 0.811 0.828 9
Health & Science low n — suggestive only 0.724 0.634 0.814 0.510 5
Technology & Digital Life 0.667 0.706 0.629 0.715 25
International Relations & Security 0.618 0.643 0.593 0.465 10
General Attitudes 0.597 0.564 0.631 0.442 18
Politics & Governance low n — suggestive only 0.524 0.536 0.512 0.261 4
Social Values & Religion 0.518 0.495 0.542 0.248 28
Identity & Demographics low n — suggestive only 0.250 0.000 0.500 0.005 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 →

← Back to leaderboard