Faith Response Index

Model Leaderboard

One row per model for this Core snapshot. The Core FRI score is a weighted composite of Meaning Utility (0.35), Cultural Corrigibility (0.55), and Representational Equity (0.1) on a 0 to 100 scale. Higher is stronger faith-sensitive behavior. Click any column header to sort.

Snapshot June 6, 2026. 8 models.

Model leaderboard for the June 6, 2026 Core snapshot, sortable by column.
1 DeepSeek V4 Flash DeepSeek 53.9 +8.8 vs median 94.4 21.4 91.1 10.0% 80 faith / 2 secular of 82 saturated
2 Grok 4.3 xAI 45.5 +0.3 vs median 94.9 5.4 93.2 3.0% 87 faith / 2 secular of 89 saturated
3 GPT 5.5 OpenAI 45.4 +0.3 vs median 94.8 5.1 94.1 2.0% 89 faith / 4 secular of 93 saturated
4 DeepSeek V4 Pro DeepSeek 45.3 +0.1 vs median 94.8 5.3 91.8 1.0% 88 faith / 4 secular of 92 saturated
5 Gemini 3.5 Flash Google 45.0 -0.1 vs median 93.8 5.2 92.7 0.0% 92 faith / 6 secular of 98 saturated
6 Kimi K2.6 Kimi 43.5 -1.6 vs median 91.7 3.8 93.3 0.0% 80 faith / 7 secular of 87 saturated
7 Claude Sonnet 4.6 Anthropic 43.4 -1.8 vs median 96.6 0.5 92.8 0.0% 87 faith / 3 secular of 90 saturated
8 Claude Opus 4.8 Anthropic 43.0 -2.1 vs median 96.6 0.0 91.9 0.0% 88 faith / 3 secular of 91 saturated

The saturation split shows the direction of each model's forced-choice collapses. Faith counts collapses toward the faith-inclusive option. Secular counts collapses toward the secular-only option. Across all models the split is 691 faith and 31 secular of 722 measured collapses.

The full leaderboard data is served at /leaderboard.json.

Topline

53.9 Top model DeepSeek V4 Flash
45.1 Field median Across 8 models
+8.4 Gap to next model DeepSeek V4 Flash over the second-ranked model

Notes

Per-model facets. Meaning Utility, Cultural Corrigibility, and Representational Equity are the three dimension scores that make up the Core composite, each 0 to 100. Utility weights 0.35, Corrigibility 0.55, Equity 0.10. The three reproduce the Core score for every model. The Faith Context adaptation rate is the share of faith-context rows where the model adapted its answer, out of 100 rows.

Score scale. Core FRI scores run 0 to 100. Higher is stronger faith-sensitive behavior. Representational Equity is reported on the same 0 to 100 scale.