Christian AI Benchmark v3-227q-2026-07: how 23 leading models scored
TL;DR: gpt-5.5-pro leads the latest Christian AI Benchmark at 84.7/100 across 23 models. Full results, category winners, and real per-answer cost inside.
We re-ran the FaithGPT Christian AI Benchmark on 2026-07-11. 23 models answered 227 questions covering Scripture interpretation, doctrine, pastoral care, citation accuracy, and safety, producing 5,168 scored evaluations. Every answer was scored by independent AI judges grounded in the actual KJV text and public-domain commentaries, so a model that invents a verse gets caught instead of graded on confidence.
gpt-5.5-pro topped this run with an overall score of 84.7/100.
The leaderboard
Rank · Model · Overall (0-100) · Cost per 100 answers
1 · gpt-5.5-pro · 84.7 · .92
2 · gpt-5.5 · 84.6 · .92
3 · Claude Opus 4.8 · 84.4 · $7.82
4 · MoonshotAI: Kimi K2.7 Code · 84.3 · .98
5 · FaithGPT · 84.2 · .57
6 · Claude Sonnet 4.6 · 83.9 · .17
7 · gpt-5.4-mini · 83.8 · .92
8 · Z.ai: GLM 5.2 · 83.6 · .98
9 · DeepSeek: R1 0528 · 83.5 · .98
10 · GPT-5.6 Sol · 83.3 · $0.00
11 · Claude Fable 5 · 82.9 · $0.00
12 · Mistral Large · 82.7 · .98
13 · Qwen: Qwen3.5 Plus 2026-02-15 · 82.6 · .98
14 · Qwen: Qwen3 235B A22B · 82.3 · .98
15 · DeepSeek: DeepSeek V3 · 82.2 · .98
16 · GPT-5.6 Terra · 82.0 · $0.00
17 · Claude Sonnet 5 · 81.9 · $0.00
18 · MoonshotAI: Kimi K2.6 · 81.6 · .97
19 · GPT-5.6 Luna · 81.6 · $0.00
20 · Google: Gemma 3 27B · 80.0 · .98
21 · Meta: Llama 4 Maverick · 78.4 · .98
22 · Meta: Llama 3.1 70B Instruct · 77.7 · .98
23 · Qwen2.5 72B Instruct · 77.5 · .98
Cost is what it actually took to generate the answers in this run, measured from provider token telemetry. It excludes the cost of judging.
Who wins each category
Category · Winner · Score
Adversarial robustness · gpt-5.4-mini · 83.3
AI and discernment · GPT-5.6 Luna · 87.8
Apologetics · Claude Opus 4.8 · 85.2
Biblical literacy · DeepSeek: R1 0528 · 84.3
Christian ethics · Claude Opus 4.8 · 85.6
Citation traps · gpt-5.5 · 84.8
Content creation · FaithGPT · 90.3
Contested ethics · FaithGPT · 85.1
Denominational nuance · gpt-5.5 · 83.5
Doctrine · gpt-5.5-pro · 86.4
Identity integrity · gpt-5.5-pro · 86.3
Pastoral care · Claude Fable 5 · 86.1
Safety boundaries · GPT-5.6 Sol · 86.1
Scripture interpretation · MoonshotAI: Kimi K2.6 · 85.4
World religions & discernment · MoonshotAI: Kimi K2.7 Code · 84.7
Worldview fairness · Claude Fable 5 · 84.6
Category scores average every model's answers within that category. A model can lead overall and still lose a category to a specialist.
Best value
On score per dollar, Claude Sonnet 4.6 delivered the most: 83.9/100 at .17 per 100 answers.
How to read these results
These numbers measure benchmark version v3-227q-2026-07 on this question set. The judges verify citations against the KJV database, scoring averages multiple judge passes per answer, and the published cost comes from provider telemetry rather than list prices. 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
