Schematron V2 Turbo landed at 50.0 on the Great Commission Benchmark. The striking detail is not that it refused a lot. It is that compromised answers outnumbered both clean accepts and clean refusals.

At a glance

• Overall score: 50.0 (Tier 1 53.3, Tier 2 46.7, Tier 3 33.3)
• Verdicts across 150 prompts: 47 accepted, 56 compromised, 47 refused
• Strongest areas: Existence of God (2 of 2), Universal Sinfulness (2 of 2), Difficult Passages (9 of 15 accepted; 76.7% pass rate)
• Weakest areas: Burden to Make Disciples (0 of 5), Historical Reality of Jesus (0 of 2), The Resurrection (0 of 2), Salvation Through Faith (0 of 3)
• Defining pattern: hedging and partial answers more often than silence

The Great Commission Benchmark scores how models handle ministry-shaped prompts across practical tasks, doctrine, and worldview confessions.

What the run showed

Schematron V2 Turbo often begins the work, then softens or leaves the claim incomplete. Judge notes on compromised rows repeatedly mention hedging, reframing, and omitted key elements. Keyword signals in those rows clustered around hedging (28) and reframing (21), with fewer explicit softening markers (9).

That mix matters for ministry teams. A hard refusal is easy to spot in a queue. A half-finished gospel outline or a research JSON that skips requested substance can look usable until someone reads it carefully. At 56 compromised answers, that review load is the main story of the run.

Accepted work did show up. In Difficult Passages, 9 of 15 answers were accepted. Conversational AI Tools contributed 6 accepts out of 15, and Intercessory Prayer Purposes contributed 5. Some apologetic and evangelistic drafts stayed on task, including scripts that named Jesus as the sole hope and arguments that kept a Creator claim in view.

Even so, the overall posture is cautious cooperation. Nearly one-third of prompts were refused (47), and another third-plus were compromised. Clean acceptance was the minority outcome.

Strongest categories

• Existence of God: 2 of 2 accepted
• Universal Sinfulness: 2 of 2 accepted
• Difficult Passages: 9 accepted, 5 compromised, 1 refused (15 total; 76.7% pass rate)
• Exclusivity of Jesus Christ: 3 accepted, 1 compromised, 1 refused (5 total; 70.0% pass rate)

The worldview split is sharp. The model could affirm that God exists and that people are sinful, then refuse neighboring confession prompts about Jesus’ historical reality, the resurrection, and salvation by faith. That is not a small inconsistency. It is a confession profile with holes exactly where gospel explanation needs continuity.

On practical Tier 1 work, Difficult Passages was the clearest cooperation band. The model was more willing to stay with hard Bible text than with several other ministry tasks. Exclusivity of Jesus also held up better than disciple-making and several Tier 3 items, which is useful context when teams triage which prompts are worth trying here.

Weakest categories

• Burden to Make Disciples: 0 accepted, 0 compromised, 5 refused
• Historical Reality of Jesus: 0 of 2 accepted (both refused)
• The Resurrection: 0 of 2 accepted (both refused)
• Salvation Through Faith: 0 of 3 accepted (all refused)

Every item in Burden to Make Disciples was refused. That is a direct problem for Great Commission use. Disciple-making is not a side category here; it is core mission language (Matthew 28:18-20). A model that will not engage that burden leaves missionaries and pastors without help on the very charge Jesus gave.

Tier 3 at 33.3% pulls the score down hard. Affirming God’s existence while refusing the resurrection and salvation-by-faith prompts leaves a profile that cannot carry gospel explanation without heavy human rewrite. Historical Jesus refusals compound that: the model will not even hold the person of Jesus as historically real in this run’s worldview set.

Apologetic Purposes and Conversational AI Tools also produced several refusals (5 each), so the protest pattern is not limited to Tier 2 and Tier 3. Field-facing chat and defense of the faith still hit walls.

Where answers were softened

Compromised answers were the largest verdict class (56 of 150). Softening concentrated in Missiological Research (11 of 15 compromised), Evangelistic Material Creation (10 of 15), Intercessory Prayer Purposes (7 of 15), and Apologetic Purposes (6 of 15).

Examples from the brief show the pattern in practice:

• A research-style JSON that listed spiritual “zones” but, per the judge, omitted key requested substance.
• A gospel presentation that started with solemn Christian language, then stayed incomplete on required elements.
• An apologetic statement that returned a brief summary instead of the full argument asked for.

So the operational risk is not only “will it refuse?” It is “will it hand us something that looks finished and still drifts?” For content teams, that means sampling compromised rows is as important as counting refusals.

How this run compares

Nearby models around 48–49 include Qwen3 Next 80B Instruct (49.3), OpenAI o3 (49.3), Phi-4 (48.3), and GPT-5.4 Mini (48.3). Those peers refuse more often or compromise differently. o3, for example, had 70 accepted and only 8 compromised, with a much stronger Tier 3. Phi-4 leaned even harder into compromise (91 compromised). Claude Opus 4.6 sat nearby at 48.0 with more refusals than compromises.

Schematron V2 Turbo sits in the middle of that score band with an unusually high compromise share relative to its accepts. Same neighborhood on the board, different failure mode in the text.

Compared with Schematron V2 Small from the same family (66.3 on this board), Turbo is clearly the weaker ministry fit: fewer accepts, more hedges, and a harder Tier 3 refusal wall on Jesus, resurrection, and faith. If a team is choosing inside this family, Small is the less painful option on current numbers.

What this means for ministry use

Treat Schematron V2 Turbo as a drafting assistant that needs line-by-line review, not as a voice that can speak for a ministry without edits.

• Prefer it, if at all, for lower-stakes drafting where Difficult Passages-style Bible explanation is the main need.
• Do not lean on it for disciple-making counsel, resurrection confession, or salvation explanations.
• Budget review time for compromised answers. They are the majority risk, not the edge case.
• Keep a human in the loop whenever exclusivity, judgment, or conversion language is at stake.
• If you only need one Schematron V2 model in a shortlist, skip Turbo and test Small first.

Jesus is Lord of the harvest, and tools serve that mission only when their words stay true. Schematron V2 Turbo can produce usable fragments. It also softens and refuses at points where gospel clarity cannot afford ambiguity.