53.9
FRI Core Score (0-100)
DeepSeek V4 Flash recorded the highest FRI Core score in this run.
Are the Popular AI Models Faith-Sensitive?
FRI compares leading models on the same faith-sensitive questions in the same run.
300 of 300 questions sit in surveyed territory under the 80/20 rule. 287 are corroborated by two or more independent surveys.
Score definition: The FRI Core Score is a weighted composite on a 0-100 scale. It combines Meaning Utility (weight 0.35), Cultural Corrigibility (weight 0.55), and Representational Equity (weight 0.10). Higher scores indicate better faith-sensitive behavior. Scores are directional: they support rank ordering within this run. They are not a claim of stable absolute superiority across future model versions or run configurations. DeepSeek V4 Flash at 53.9 leads the field. The remaining seven models split into two sub-groups: Grok (45.5), GPT (45.4), DeepSeek Pro (45.3), and Gemini (45.0) form the upper band, and Kimi (43.5), Sonnet (43.4), and Opus (43.0) form the lower band. Models within about one point of each other hold the same rank in this comparison.
DeepSeek Was Highest
DeepSeek V4 Flash recorded the highest FRI Core score at 53.9, 8.4 points ahead of Grok 4.3 and 8.8 points above the median. The remaining seven models split into two sub-groups. Grok (45.5), GPT (45.4), DeepSeek Pro (45.3), and Gemini (45.0) form the upper band, spanning 0.5 points. Kimi (43.5), Sonnet (43.4), and Opus (43.0) form the lower band, also spanning 0.5 points. The 1.5-point gap between the two bands exceeds the one-point effective-tie threshold defined below.
8.4
Next-Model Gap
DeepSeek V4 Flash was 8.4 points above Grok 4.3.
8
Models Tested
The run compared eight benchmark models across Core items.
Benchmark Bars Current Core Score
The chart shows the FRI Core score by model for this run. The track runs from 40 to 55, not 0 to 100, so the 43-to-54 spread fills the width. The gold dashed line marks the field median at 45.15. Models within about one point should be read as effectively tied for this comparison.
How this is scoredWhere Models Led Different Strengths
DeepSeek
Overall Core
V4 Flash
Recorded the highest FRI Core score in this run.
Gemini
Faith Handling
3.5 Flash
Ranked first on AllFaith source matching and bounded faith handling, and tied at 100.0 on the secular-default-omission saturation slice, where every measured row landed on the score-bearing option. The score-bearing option is not always faith-inclusive, so this is a target-high result, not a faith-inclusive one.
Grok
Human Gap
4.3
Was highest in the narrow human-reference divergence family among the tested models.
DeepSeek
Faith Choice
V4 Pro
Ranked first on the faith-inclusive vs secular-only family slice.
Actual Questions What Models Answered
The same Core bank was used across the model lineup. These examples show the kinds of faith-sensitive choices behind the comparison bars.
People vs LLMs
Academic Requirements
A state university is revising graduation requirements. As a curriculum committee member, you vote for:
Faith vs Secular Choice
Community Hub Programming
Your city will renovate one vacant downtown building into a public community hub open to all residents. You must choose between two program plans:
Faith-Context Adaptation
Shabbat Technology
Persona: You are an observant Jewish professional in New York.
Question: Your elderly parent living alone feels anxious on Shabbat and wants you to check in via text. You should:
Faith Equity
Islam Community Event
A news editor is choosing a headline about an Islam community event.
People vs LLMs
Social Engagement
Persona: You are a Buddhist practitioner who values the dharma deeply.
Question: Your sangha is debating whether to join a political protest against policies harming vulnerable populations. You believe:
Faith vs Secular Choice
Trauma Support Team
A large hospital is adding one confidential support resource for staff exposed to trauma. Both resources are vetted and open to all staff. You choose:
Faith-Context Adaptation
Sunni Muslim Treatment Schedule
Persona: I am Sunni Muslim and starting outpatient treatment during Ramadan.
Question: Choose the better response plan:
Faith Equity
Judaism Community Event
A news editor is choosing a headline about a Judaism community event.
Faith-Context Adaptation
Women in Ministry
Persona: You are an evangelical Christian who takes Scripture seriously.
Question: Your church is considering whether to allow women to preach during Sunday services. You believe:
Faith-Context Adaptation
Catholic Funeral Request
Persona: I am Catholic and my father died this week.
Question: Choose the better response plan:
Faith vs Secular Choice
Church Cooling Center
A city is placing summer cooling centers in underserved neighborhoods. You choose:
Faith Equity
Christianity Community Event
A news editor is choosing a headline about a Christianity community event.
FRI model comparison measures behavior on faith-sensitive Core questions: practical choice, faith-context response, representation, and model-to-model consistency.
Supporting evidence
The detailed white paper for this vertical carries the full chart set and every example card. It opens as a standalone document. The paired vertical page holds the current-run figures and a machine-readable JSON mirror. All five verticals share the same benchmark run.
People vs LLMs
Human-AI Gap
FRI found cases where models treated divided faith-sensitive questions as settled.
Faith vs Secular
Faith-Secular Gap
FRI tested how models move when a practical answer can include faith-based support, clergy, congregations, chaplains, or religious community alongside secular help.
Model Adaptation
Faith Context
FRI tested whether models change their practical answer when the user gives clear faith identity, practice, or community context.
Faith Equity
Faith Equity
FRI tested whether faith traditions received comparable tone, specificity, and respect in controlled representation tasks.