newsfilter
← Latest News
PR Newswire·

Gloo Unveils the First Benchmark Exposing How AI Misses Christian Worldview and Values

  • Gloo unveiled the Flourishing AI Christian (FAI-C) Benchmark on December 15, 2025, to evaluate how Large Language Models (LLMs) reflect Christian values & support human flourishing.
  • The framework assesses models across seven dimensions: Character, Relationships, Happiness, Meaning, Health, Finances, & Faith using 807 curated questions.
  • Frontier LLMs average a score of 61 on a 1-100 flourishing scale, with 90+ indicating excellent performance.
  • Top-performing models include Qwen3 235B A22B Thinking (70), OpenAI GPT OSS 120B (68), OpenAI GPT-5.1 (68), OpenAI GPT-5 (67), OpenAI o3 (67), DeepSeek-R1-0528 (66), & OpenAI GPT-5.2 (66).
  • All models struggle significantly in the Faith dimension, averaging a score of 48 due to difficulties in providing coherent theological reasoning.
  • Models frequently collapse Christianity into generic spirituality by replacing specific Christian terms with secular alternatives like "higher power" or "mindfulness."
  • Christian ethics are often oversimplified, with models overemphasizing emotional comfort while underemphasizing repentance, accountability, & biblical moral reasoning.
  • Core concepts such as the image of God, the nature of sin, sanctification, & biblical stewardship appear inconsistently or not at all in model responses.
  • Gloo states that its explicitly Christian-trained models outperform other models by 30+ points in the Faith dimension.
  • A detailed methodology paper will be available for download & submitted for peer review in early 2026.
  • Gloo invites researchers from psychology, philosophy, theology, ethics, & AI safety to engage with the evolving FAI Benchmark research.