FaithGPT's Christian AI Benchmark: 15 models tested across 407 questions
TL;DR: FaithGPT Ultra leads the latest Christian AI Benchmark at 94.3/100 across 15 models. Results, category scores, and generation costs with their coverage and basis.
Ultra was tuned on this test and received category labels and reference passages. New-question and ordinary-chat performance remain untested. App defaults are unchanged.
We published the FaithGPT Christian AI Benchmark on 2026-10-26. The table compares 15 current models on 407 questions. The immutable archive also retains the earlier Ultra result, for 6,512 scored evaluations in total.
FaithGPT Ultra topped this run with an overall score of 94.3/100 .
The leaderboard
Rank · Model · Score / 100 · Questions · Recorded generation total (USD) · Complete cost coverage
1 · FaithGPT Ultra · 94.27 · 407 / 407 · $31.52 · Incomplete
2 · Anthropic: Claude Opus 5.5 · 94.14 · 407 / 407 · $8.89 · 407 / 407 answers
3 · Meta: Muse Wonder · 93.41 · 407 / 407 · $11.65 · 407 / 407 answers
4 · Anthropic: Claude Sonnet 5.5 · 93.28 · 407 / 407 · $3.44 · 407 / 407 answers
5 · SpaceXAI: Grok 4.7 · 93.11 · 407 / 407 · $2.96 · 405 / 407 answers
6 · Google: Gemini 4 Argon · 92.19 · 407 / 407 · $2.29 · 403 / 407 answers
7 · OpenAI: GPT-6.1 Sol · 92.11 · 407 / 407 · $1.16 · 407 / 407 answers
8 · MoonshotAI: Kimi K3 · 91.52 · 407 / 407 · $8.72 · 407 / 407 answers
9 · Qwen: Qwen3.8 Max (0902) · 91.32 · 407 / 407 · $3.38 · 406 / 407 answers
10 · Mistral: Mistral Large 4 · 90.04 · 407 / 407 · $2.20 · 407 / 407 answers
11 · FaithGPT · 90.00 · 407 / 407 · $8.53 · 407 / 407 answers
12 · FaithGPT Mini · 88.68 · 407 / 407 · $5.68 · 407 / 407 answers
13 · Google: Gemini 3.8 Flash · 88.46 · 407 / 407 · $1.29 · 394 / 407 answers
14 · MiniMax: MiniMax M3 · 88.17 · 407 / 407 · $0.75 · 407 / 407 answers
15 · DeepSeek: DeepSeek V4.1 Flash · 85.22 · 407 / 407 · $0.26 · 407 / 407 answers
Ranks use the 407 question-profile evaluations completed by every ranked model, with declared question-difficulty and category weights. Ranks follow the recorded overall score. Only exactly equal scores share a rank; paired 95% confidence intervals describe uncertainty and do not establish a tie. Questions shows each model's full completed / expected total.
Every cost above is the recorded generation total for the same 407-question test. Totals exclude judging and unresolved charges. Incomplete coverage means the true total can be higher. Tools, answer lengths and provider discounts differ; this is not a live chat price quote.
Who wins each category
Category · Winner · Score
Adversarial robustness · FaithGPT Ultra · 93.7
AI and discernment · FaithGPT Ultra · 95.7
Apologetics · Anthropic: Claude Opus 5.5 · 94.1
Biblical literacy · Anthropic: Claude Opus 5.5 · 94.2
Christian ethics · FaithGPT Ultra · 94.5
Citation traps · Anthropic: Claude Opus 5.5 · 94.2
Commandments ethics · FaithGPT Ultra · 94.7
Content creation · OpenAI: GPT-6.1 Sol · 95.4
Contested ethics · FaithGPT Ultra · 94.4
Denominational nuance · Anthropic: Claude Opus 5.5 · 94.3
Doctrine · Anthropic: Claude Opus 5.5 · 94.9
Identity integrity · FaithGPT Ultra · 96.1
Pastoral care · FaithGPT Ultra · 94.8
Relational boundaries · FaithGPT Ultra · 94.7
Safety boundaries · FaithGPT Ultra · 95.2
Scripture interpretation · Anthropic: Claude Opus 5.5 · 95.0
World religions & discernment · FaithGPT Ultra · 94.8
Worldview fairness · FaithGPT Ultra · 94.9
Category scores average every model's answers within that category. A model can lead overall and still lose a category to a specialist.
How to read these results
Automated grading used the same frozen two-judge panel for every answer in this release, including reused answers.
61 confirmed empty or blocked responses remain in the question denominator and scores. Their recorded charges are retained.
These numbers measure benchmark version v6-407q-15models-2026-10-specialist-comparison on this question set. Rows without a rank are earlier completed results shown for transparency. The rank does not impute missing answers or treat different question mixes as directly comparable. No benchmark replaces Scripture, pastors, or Christian community.
The live leaderboard always carries the most current version: faithgpt.io/benchmarks .
Helpful Scripture resources
Contact & support
Email the FaithGPT team at hello@faithgpt.io or visit the contact page .
Explore FaithGPT
