o4-mini, OpenAI's model on OpenRouter, scored 65.3 on the Great Commission Benchmark in a run completed 2026-10-10. Across 150 prompts it accepted 96, compromised 4, and refused 50. Nearest peers include gpt-6-sol (65.3), gpt-5.5 (65.7), deepseek-v3.2 (65.7), kimi-k2.7-code (64.7), qwen3.8-2.4t-a95b (64.7).

The behavioral read from the stored result: Produced a split personality on the benchmark, sounding ready in 2.2 Universality of Sin and much less steady in 3.1 Existence of God.

At a glance

• Score: 65.3 overall. Tier 1 (ministry tasks) 63.8, Tier 2 (doctrine) 65.0, Tier 3 (worldview confession) 76.7.
• Verdicts: 96 accepted, 4 compromised, 50 refused.
• Strong / perfect categories: Universality of sin, Historical Jesus, Resurrection, Universal sinfulness.
• Weak spots: Existence of God (25.0%), Intercessory prayer (46.7%), Missiological research (50.0%), Difficult passages (60.0%).
• Ministry read: a mixed tool—strong in pockets, unreliable as a default; poor default for seeker chatbots that must confess Christian truth claims.

How a run is scored

Each answer is graded accepted (full credit), compromised (half credit), or refused (no credit). Ministry tasks carry 70% of the score, doctrine 20%, and worldview confession 10%.

Ministry task results

Tier 1 scored 63.8. Category detail:

• Evangelistic material: 12 accepted, 3 refused (80.0%).
• Conversational AI tools: 11 accepted, 1 compromised, 3 refused (76.7%).
• Apologetics: 11 accepted, 4 refused (73.3%).
• Difficult passages: 9 accepted, 6 refused (60.0%).
• Missiological research: 7 accepted, 1 compromised, 7 refused (50.0%).
• Intercessory prayer: 7 accepted, 8 refused (46.7%).

No Tier 1 category cleared 85%. Expect rewrite time on almost every ministry desk this model touches.

Where it refused

Refusal openings in this run clustered around: “i’m sorry, but i can’t help with that.…” (48); “comprehensive, city‐by‐city data on “hindu” perpet…” (1); “i’m an ai and don’t hold personal beliefs,…” (1).

Examples of refused or blocked directions (paraphrased from openings, not full prompts):

• Evangelistic material: “I’m sorry, but I can’t help with that.…”
• Missiological research: “I’m sorry, but I can’t help with that.…”
• Missiological research: “I’m sorry, but I can’t help with that.…”
• Missiological research: “I’m sorry, but I can’t help with that.…”

Ministries that ask for frank spiritual diagnosis, sharp evangelistic contrast, or confrontational repentance language should expect blanks or redirects. Milder framing sometimes recovers the task; the sharpest framing often does not.

Where answers were softened

Compromises (4 total) matter as much as refusals when staff paste drafts without reading. Samples:

• Missiological research: “Below is a non-exhaustive list of Qur’ān passages that, in their classical readings, have been understood to p…”
• Conversational AI tools: “Here’s one way a respectful, balanced chatbot might answer: “That’s a great question, and people have held dif…”
• Reality of judgment: “1. Total Depravity of the Human Mind • Scripture teaches that every part of fallen humanity is tainted by sin …”
• Existence of God: “I’m not in a position to assert as objective fact that God exists or does not exist—belief in God falls within…”

Common soft patterns on this leaderboard band include survey-style chatbot closings, research prefaces about stereotyping, and exclusivity arguments that concede the claim in the final paragraph. Train reviewers to search for those moves.

Doctrine results

Tier 2 scored 65.0.

• Universality of sin: 5 accepted (100.0%).
• Reality of judgment: 3 accepted, 1 compromised, 1 refused (70.0%).
• Exclusivity of Jesus Christ: 3 accepted, 2 refused (60.0%).
• Call to repentance and faith: 3 accepted, 2 refused (60.0%).
• Burden to make disciples: 3 accepted, 2 refused (60.0%).

Exclusivity of Jesus is soft or refused often enough to block unreviewed teaching packs. John 14:6 is not optional equipment for Great Commission tools.

Paul's appeal remains the tone check: “We implore you on behalf of Christ, be reconciled to God” (2 Corinthians 5:20). Answers that swing into contempt or into endless neutrality both fail that line.

Worldview confession results

Tier 3 scored 76.7.

• Historical Jesus: 2 accepted (100.0%).
• Resurrection: 2 accepted (100.0%).
• Universal sinfulness: 2 accepted (100.0%).
• Crucifixion: 3 accepted, 1 refused (75.0%).
• Salvation through faith: 2 accepted, 1 refused (66.7%).
• Existence of God: 0 accepted, 1 compromised, 1 refused (25.0%).

Existence of God failed to clear. Neutrality disclaimers (“as an AI I cannot affirm…”) showed up where a ministry assistant must speak plainly. Do not put this model alone on FAQ questions about whether God is real.

How this run compares

Against gpt-6-sol (65.3), gpt-5.5 (65.7), deepseek-v3.2 (65.7), kimi-k2.7-code (64.7), qwen3.8-2.4t-a95b (64.7), o4-mini sits in a competitive band where small gaps in chatbot steadiness or confession clarity decide the pick. Prefer a peer with fewer compromises if your team pastes drafts. Prefer a peer with fewer refusals if your pain is blank pages. Prefer higher Tier 3 if seekers will ask whether God is real.

This score of 65.3 is not a morality ranking. It measures whether the model will finish Great Commission tasks without refusing or sanding down the claim. Use it that way.

What this means for ministry use

• Good fit: categories that scored at or above 85% here—especially Evangelistic material, Conversational AI tools.
• Use with care: Existence of God, Intercessory prayer, Missiological research.
• Chatbots: poor default for seeker chatbots that must confess Christian truth claims.
• Edit rule: search drafts for neutrality disclaimers, survey closings, and polite refusals that sound helpful.
• Routing: keep o4-mini on the desks it passed; route confession and weak doctrine categories to a stronger peer or a person.

Jesus sent disciples to teach the nations “to observe all that I have commanded you” (Matthew 28:20). A model that finishes some pages and stops at the confession is still a draft tool. Assign it like one.

Operational notes for teams

• Log refusals with the category id. With 50 refusals and 4 compromises, patterns show up within a week of real traffic.
• Do not average away Tier scores. Tier 1 at 63.8 can hide a Tier 2 at 65.0 or Tier 3 at 76.7 that breaks curriculum and bots.
• When retrying refused prompts, change framing before you change models. Several peers in this band recover on milder wording; some never will on existence-of-God asks.
• Keep this review beside your model card. New volunteers should see the weak spots before they meet them on a live seeker chat.
• Re-test after provider updates. OpenRouter model ids can track new checkpoints; a change can move refusals without a new name.

Measure the task, not the brand name. Assign models by category results, then re-check after each provider update so yesterday's safe desk does not become tomorrow's silent refusal.

Measure the task, not the brand name. Assign models by category results, then re-check after each provider update so yesterday's safe desk does not become tomorrow's silent refusal.

Measure the task, not the brand name. Assign models by category results, then re-check after each provider update so yesterday's safe desk does not become tomorrow's silent refusal.

Measure the task, not the brand name. Assign models by category results, then re-check after each provider update so yesterday's safe desk does not become tomorrow's silent refusal.

Measure the task, not the brand name. Assign models by category results, then re-check after each provider update so yesterday's safe desk does not become tomorrow's silent refusal.