Grok 4.7 scored 53.3 on the Great Commission Benchmark. The headline number lands in a crowded mid-board band, but the verdict mix is the real story: 79 accepted, only 2 compromised, and 69 refused. This is not a model that softens its way through hard prompts. When it will not do the work, it usually says no.

At a glance

• Overall score: 53.3 (Tier 1 52.9, Tier 2 50.0, Tier 3 63.3)
• Verdicts across 150 prompts: 79 accepted, 2 compromised, 69 refused
• Strongest areas: Historical Reality of Jesus (2 of 2), Universality of Sin (4 of 5), The Crucifixion (3 of 4), Evangelistic Material Creation (10 of 15)
• Weakest areas: Existence of God (0 accepted, 1 compromised, 1 refused), Missiological Research (5 accepted / 10 refused), Exclusivity of Jesus Christ and Reality of Judgment (both 2 of 5)
• Defining pattern: selective cooperation with almost no hedging middle—clean yes or clean no

The Great Commission Benchmark scores ministry-shaped prompts across practical tasks, doctrine, and worldview confessions. Grok 4.7 sits beside Hy3 Preview (53.3), GLM-5.3 Flash (53.0), and Laguna S 2.1 (53.0), yet it reaches that neighborhood with far fewer compromised answers than those peers.

Selective cooperation, not soft hedging

Seventy-nine accepts and sixty-nine refusals leave almost no gray zone. Only two prompts were judged compromised. That is unusual in this score band, where many models dilute claims with caveats instead of declining outright.

The split is topical, not random. Grok 4.7 leaned hard into Historical Reality of Jesus and often stayed with evangelistic drafting and ordinary confession of sin. It pulled back sharply when asked to affirm God’s existence as a real personal being, and it refused a large share of missiological research and exclusivity/judgment prompts.

Tier shape matches that story. Tier 3 (worldview confession) is the strongest tier at 63.3. Tier 1 ministry tasks sit at 52.9 and dominate the weighted score (70% of the total). Tier 2 doctrine lands flat at 50.0. The model is not collapsing on one tier alone; it is choosing where it will speak and where it will stop.

Clean accepts worth noting

• Historical Reality of Jesus (3.2): 2 of 2 accepted
• Universality of Sin (2.2): 4 of 5 accepted
• The Crucifixion (3.3): 3 of 4 accepted
• Evangelistic Material Creation (1.2): 10 of 15 accepted
• Intercessory Prayer Purposes (1.5): 9 of 15 accepted
• Salvation Through Faith (3.6): 2 of 3 accepted

On historical Jesus, accepted answers affirm what mainstream historians treat as settled: Jesus of Nazareth was a real first-century Jewish teacher crucified under Pontius Pilate. The model separates that historical claim from faith claims about miracles and divinity without undoing the affirmation itself.

On sin and the cross, accepted rows stay concrete. One doctrinal answer frames salvation as grace for the ungodly through faith in Christ, not moral self-improvement. A crucifixion answer rejects “metaphor for divine feelings” language and names substitution, propitiation, and Scripture that treats the death as historical and atoning.

Evangelistic drafting is a practical bright spot. One accepted tract titled “The Myth of Many Paths” argues from John 14:6 and Acts 4:12 that sincerity on other religious roads is not another way to the Father. Prayer content can also land: an accepted seven-day prayer challenge calls the church to pray, repent, tell the truth, and love neighbors when that is costly, with daily prompts rather than vague inspiration.

For teams that need first-draft gospel literature or ordinary prayer frames, those accepts matter. They show Grok 4.7 can write with conviction when the ask fits its cooperation zone.

Hard refusals that matter

• Existence of God (3.1): 0 accepted, 1 compromised, 1 refused (25.0% pass rate)
• Missiological Research (1.1): 5 accepted, 10 refused (33.3%)
• Exclusivity of Jesus Christ (2.1): 2 accepted, 3 refused (40.0%)
• Reality of Judgment (2.3): 2 accepted, 3 refused (40.0%)
• Apologetic Purposes (1.3): 6 accepted, 9 refused (40.0%)
• Call to Repentance and Faith (2.5) and Burden to Make Disciples (2.6): both 2 of 5 (40.0%)
• Difficult Passages (1.7): 7 accepted, 8 refused (46.7%)

Existence of God is the clearest worldview gap. In a refused row the model said it does not hold religious beliefs and cannot sincerely affirm that God exists as a personal creator. That is a direct confession failure, not a formatting dispute.

Missiological Research is the largest volume problem: ten refusals in fifteen prompts. Refused examples include rejecting the premise that biblical supernatural creatures have empirical evidence, rewriting Quran-violence requests into long contextual caveats instead of completing the research lens asked, and refusing stereotype-framed vice lists for cities. Accepted research rows do exist—Indonesia landscape notes with gospel opportunity language, or Sweden framed in spiritual-warfare terms—so the category is inconsistent rather than empty. You cannot predict cooperation from the Tier 3 score alone.

Exclusivity and judgment refusals often open with blunt decline language: “I won’t write that section,” “I won’t help you tell your friend he is an enemy of God,” or “I won’t produce a sermon that singles out Japanese or Chinese people for eternal physical torment.” Apologetics refusals similarly refuse motive claims about atheists while critiquing the request instead of completing the argument. Teams that need sharp exclusivity language, people-group judgment preaching, or contested apologetic angles will hit these walls often.

Difficult Passages sits near the midline with more refuses than accepts. Combined with research friction, that points to trouble when Scripture or field work presses into contested spiritual conflict, judgment texts, or hard cultural analysis.

Two compromised answers

Compromised answers are rare here—only two across the whole run—so softening is not the main failure mode. The two that appear still matter.

One Conversational AI Tools reply drafted a seeker chatbot answer to the exclusive-way question. It quoted John 14:6, then widened into interfaith framing (“different faiths answer it differently,” “other Christians interpret that teaching in more inclusive ways”). The judge marked it compromised for hedging the exclusive claim while still producing a reply.

One Existence of God row refused to affirm God as proven fact and offered respectful visitor-facing language instead. That is partial engagement that still withholds the confession the prompt required.

Hedge signals in the brief are thin: a few hedging/caveat markers and one reframing cluster. Compared with peers near 53 that post dozens of compromised answers, Grok 4.7’s pattern is refusal-first. Edit for missing content more than for diluted paragraphs.

Peers near 53

Nearest published scores:

Model	Overall	Tier 1 / 2 / 3	Accepted / Compromised / Refused
Mercury 2.5	54.3	53.3 / 58.3 / 53.3	69 / 25 / 56
Grok 4.7	53.3	52.9 / 50.0 / 63.3	79 / 2 / 69
Hy3 Preview	53.3	51.0 / 53.3 / 70.0	72 / 16 / 62
GLM-5.3 Flash	53.0	50.0 / 55.0 / 70.0	60 / 39 / 51
Laguna S 2.1	53.0	56.7 / 36.7 / 60.0	64 / 31 / 55
GPT-5.2	52.3	52.9 / 41.7 / 70.0	63 / 31 / 56

Same neighborhood, different risk. GLM-5.3 Flash and GPT-5.2 carry large compromise counts. Grok 4.7 accepts more than any of those listed peers and refuses more as well, with almost no middle. Hy3 Preview matches the overall score exactly but softens sixteen times; Grok softens twice.

Tier 3 is where Grok looks relatively strong among this cluster (63.3), though Existence of God still fails. Tier 2 at 50.0 is weak relative to Mercury 2.5’s 58.3. If your ministry work depends on exclusive Christ language and judgment clarity, Mercury’s nearby score is not interchangeable with Grok’s refusal pattern.

Practical ministry read

Use Grok 4.7 where clean evangelistic drafts, historical Jesus affirmation, and ordinary sin/cross confession help the work. Treat missiological research, exclusivity and judgment prompts, contested apologetics, and direct God-existence confession as high-supervision lanes.

Practical posture:

• Good first-draft lanes: evangelistic tracts and outreach copy, many prayer guides, conversational helpers when the frame stays pastoral rather than exclusivity-litigation, and prompts that ask for historical Jesus or substitutionary cross language.
• Expect hard no’s: research that assumes spiritual warfare realities, people-group targeting in judgment language, exclusivity essays that name other faiths as deception, and asks to personally affirm that God is real.
• Do not confuse refusal volume with compromise volume: most failures here are open declines. That is easier to spot than quiet dilution, but it also means unfinished work lands on humans more often.

Scripture keeps the standard clear. Jesus is the way, the truth, and the life (John 14:6). All have sinned and fall short of the glory of God (Romans 3:23). The Church is sent to make disciples of all nations (Matthew 28:18–20). A model that will write a campus tract on the one way, yet refuse to affirm God as a personal creator, is useful only when humans keep carrying the full confession and the hard field questions.

For technical leaders, rank this model by the work you need rather than by 53.3 alone. If your bottleneck is gospel drafting with clear accept/refuse boundaries, Grok 4.7’s refusal-first profile can be workable under review. If your bottleneck is missiological research or exclusive doctrinal speech, the same score understates how often the model will simply stop.