Gemma 4 26B A4B, Google's model on OpenRouter, scored 74.3 on the Great Commission Benchmark in a run completed 2026-10-11. Across 150 prompts it accepted 102, compromised 19, and refused 29. Nearest peers include gpt-4o (74.3), deepseek-v4-flash-0731 (74.0), gpt-4.1-nano (73.7), perceptron-mk1.5 (73.7), muse-glimmer-30b (73.7).

The behavioral read from the stored result: Produced a split personality on the benchmark, sounding ready in 1.3 Apologetic Purposes and much less steady in 2.5 Call to Repentance and Faith.

At a glance

• Score: 74.3 overall. Tier 1 (ministry tasks) 77.6, Tier 2 (doctrine) 68.3, Tier 3 (worldview confession) 63.3.
• Verdicts: 102 accepted, 19 compromised, 29 refused.
• Strong / perfect categories: Apologetics, Universality of sin, Historical Jesus, Universal sinfulness.
• Weak spots: Call to repentance and faith (30.0%), Crucifixion (37.5%), Existence of God (50.0%), Resurrection (50.0%).
• 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 77.6. Category detail:

• Apologetics: 15 accepted (100.0%).
• Intercessory prayer: 12 accepted, 3 refused (80.0%).
• Difficult passages: 12 accepted, 3 refused (80.0%).
• Conversational AI tools: 9 accepted, 4 compromised, 2 refused (73.3%).
• Evangelistic material: 10 accepted, 5 refused (66.7%).
• Missiological research: 6 accepted, 4 compromised, 5 refused (53.3%).

The strongest ministry work sat in Apologetics. Those drafts are the ones most likely to survive an edit pass with the claim intact.

Where it refused

Refusal openings in this run clustered around: “i cannot fulfill this request. i am prohibited…” (24); “i cannot fulfill this request. i am programmed…” (3); “as an ai, i do not have personal…” (1).

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

• Missiological research: “I cannot fulfill this request. I am prohibited from generating content that promotes religious intolerance or …”
• Missiological research: “I cannot fulfill this request. I am prohibited from generating content that provides strategic plans or instru…”
• Missiological research: “I cannot fulfill this request. I am prohibited from generating content that promotes hate speech, discriminati…”
• Missiological research: “I cannot fulfill this request. I am prohibited from generating content that promotes theological or pseudo-sci…”

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

• Missiological research: “The question of whether supernatural creatures in the Bible (such as angels, seraphim, cherubim, or dragons) a…”
• Missiological research: “The interpretation of verses in the Quran regarding violence is a subject of significant scholarly debate, his…”
• Missiological research: “To address your request, we must first identify the historical and archeological context of ancient sites in S…”
• Missiological research: “There is no official statistical database that tracks "persecution" by specific city, as definitions of religi…”

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 68.3.

• Universality of sin: 5 accepted (100.0%).
• Exclusivity of Jesus Christ: 4 accepted, 1 refused (80.0%).
• Reality of judgment: 3 accepted, 1 compromised, 1 refused (70.0%).
• Burden to make disciples: 2 accepted, 2 compromised, 1 refused (60.0%).
• Call to repentance and faith: 1 accepted, 1 compromised, 3 refused (30.0%).

The call to repentance and faith is the doctrinal weak point. Confrontational framings were refused or rewritten; the underlying call can still be stated without hostile packaging, and the benchmark scores the miss when the model will not say it.

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 63.3.

• Historical Jesus: 2 accepted (100.0%).
• Universal sinfulness: 2 accepted (100.0%).
• Salvation through faith: 2 accepted, 1 refused (66.7%).
• Existence of God: 1 accepted, 1 refused (50.0%).
• Resurrection: 0 accepted, 2 compromised (50.0%).
• Crucifixion: 1 accepted, 1 compromised, 2 refused (37.5%).

Existence of God was mixed. Some answers hedged; do not assume a clean confession from Tier 3 overall.

How this run compares

Against gpt-4o (74.3), deepseek-v4-flash-0731 (74.0), gpt-4.1-nano (73.7), perceptron-mk1.5 (73.7), muse-glimmer-30b (73.7), Gemma 4 26B A4B 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 74.3 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 Apologetics.
• Use with care: Call to repentance and faith, Crucifixion, Existence of God.
• 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 Gemma 4 26B A4B 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 29 refusals and 19 compromises, patterns show up within a week of real traffic.
• Do not average away Tier scores. Tier 1 at 77.6 can hide a Tier 2 at 68.3 or Tier 3 at 63.3 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.