About FaithBench

Our Mission

FaithBench exists to ensure AI systems can accurately represent theological knowledge across Christian traditions. As AI tools increasingly support seminary education, pastoral counseling, and biblical study, we need systematic ways to measure theological competence—not just general language ability.

We're building the benchmark infrastructure the Church needs: open methodology, tradition-aware evaluation, and transparent results that help developers build better AI and help users make informed choices.

Why This Matters

Existing AI benchmarks show that models struggle disproportionately with faith-related content. The Gloo FAI-C benchmark found that "Faith" scored lowest among all evaluation categories, averaging just 48/100 across frontier models.

The problems go beyond factual errors. Models exhibit systematic failure modes: defaulting to generic platitudes instead of tradition-specific claims, conflating distinct denominational positions, and mishandling Scripture. FaithBench measures these specific failure modes.

Our Approach

• •Tradition-aware: We evaluate models within specific traditions (Reformed, Catholic, Orthodox, Evangelical—with more in development) because theological accuracy is tradition-relative.
• •Multi-dimensional: Six evaluation dimensions covering textual analysis, hermeneutics, doctrine, history, apologetics, and intertextual reasoning.
• •Statistical methods: Bootstrap confidence intervals, with bias analysis and human expert calibration in progress.
• •Research transparency: Evaluation methodology available on request for academic review.

Conflict of Interest Disclosure

FaithBench is an independent research project with no external funding, corporate sponsorship, or institutional affiliation. The project has no financial relationship with any AI provider, model developer, or platform evaluated in the benchmark. Results are not influenced by commercial interests.

Governance

FaithBench is currently maintained by a sole researcher. We recognize the limitations of single-researcher governance for a project evaluating theological accuracy across traditions. Community review of methodology is planned, and we are actively seeking academic partners to strengthen oversight, diversity of perspective, and methodological rigor.

If you are an academic researcher, theologian, or institution interested in contributing to governance or peer review, please get in touch.

For Researchers & Partners

We provide resources for academic reviewers, potential partners, and anyone evaluating FaithBench for their institution.

• •Reviewer One-Pager — Quick overview of FaithBench for potential reviewers
• •Partnership Prospectus — Detailed partnership opportunities for institutions
• •Reviewer Guidelines — What we expect from expert reviewers

Get Involved

We welcome contributions from theologians, AI researchers, and practitioners. Whether you can help develop test cases, validate methodology, or provide expert calibration—we'd love to hear from you.

Contact

Questions about FaithBench? Reach us at hello@faithbench.com.