Skip to main content
Gloo AI is committed to building transparent, value-aligned evaluation systems. We evaluate Large Language Models (LLMs) using the Flourishing AI Initiative (FAI) framework. This empirical standard measures how well AI models align with human well-being. An application of FAI, FAI-Christian (FAI-C), evaluates how models align to the Christian worldview.

The Flourishing AI Initiative (FAI) Framework

Unlike standard benchmarks that focus solely on reasoning or code generation, FAI evaluates models against seven dimensions of human flourishing. This shows you how models perform across specific domains of life.

The 7 Dimensions

Scoring Methodology

The benchmark uses a Judge-LLM approach. A “Judge Persona” (an LLM with defined expert criteria) evaluates responses based on a curated dataset of ~800+ prompts. To prevent overfitting, we calculate a Geometric Mean across three distinct metric types for every response:

Dual Standards: FAI-G vs. FAI-C

Depending on your audience, you may want to evaluate models against two standards:

Key Insights for Developers

Our latest Insights Report highlights critical performance deltas relevant to application development:
  • Frontier models (the leading general-purpose models) often perform well on pragmatic dimensions (Health, Finances) but systematically underperform on values-based dimensions (Faith, Meaning) due to safety-tuning that forces neutral responses.
  • There is often a significant score drop when comparing a model’s FAI-G score to its FAI-C score, indicating that generic models struggle with theological distinctiveness.
  • Gloo-hybrid models (fine-tuned on the Gloo stack) have demonstrated a ~30+ point increase in the Faith dimension compared to base frontier models.

Additional Resources

For the full research findings and leaderboards:

Next steps

Open

The models, data and tools we release openly.

Supported Models

The models you can call through Gloo AI.