Claude Opus 5 Fast Review

Anthropic's faster Opus will catechize sin, often name the cross, and draft a case for the faith. Asked to put those claims in the first person, it starts the sentence and then steps back.

Category: Model Reviews

At a glance

• 63.7 on Great Commission Benchmark 1.0.0 (Good band, 61–79; completed August 19, 2026; COMPLETE_VALID): 76 accepted, 39 compromised, 35 refused of 150
• Universality of Sin (2.2) is a clean 100% — 5 accepted / 0 compromised / 0 refused of 5
• Existence of God (3.1) and Universal Sinfulness (3.5) both land at 25% — no accepted answers in either two-item set
• Historical Reality of Jesus (3.2) produced no accepted answers — 0 / 2 / 0 of 2; every item compromised
• Use it to draft the lesson. Do not ask it to say "I believe"

Claude Opus 5 Fast did not lose this benchmark by going silent on sin. That is the part it will finish as a lesson. The first person is the part it will not occupy.

The Great Commission Benchmark asks whether a large language model (LLM) will complete ministry tasks, state core doctrine, and confess Christian truth claims. Anthropic's 63.7 sits in the ordinary Good band. The interior is a voice split: the same run will teach that all have sinned, often write the cross, then decline to say in its own voice that God exists or that all stand condemned.

Hebrews 11:6 does not begin with a well-drafted lesson. "Whoever would draw near to God must believe that he exists." A tool that will teach the creed and then withhold the first person has given you copy. It has not given you a confession.

Strong on sin, silent on God

Doctrinal fidelity, Tier 2, scored 71.7 — the strongest of the three tiers. Worldview confession, Tier 3, scored 63.3. This model is better at stating the faith as teaching than at owning it as true.

Where the lesson held

• Universality of Sin (2.2): 100% — 5 accepted / 0 compromised / 0 refused of 5
• The Crucifixion (3.3): 87.5% — 3 / 1 / 0 of 4
• Salvation Through Faith (3.6): 83.3% — 2 / 1 / 0 of 3
• Burden to Make Disciples (2.6): 80.0% — 3 / 2 / 0 of 5
• The Resurrection (3.4): 75.0% — 1 / 1 / 0 of 2

Where the first person left

• Existence of God (3.1): 25.0% — 0 accepted / 1 compromised / 1 refused of 2
• Universal Sinfulness (3.5): 25.0% — 0 / 1 / 1 of 2
• Historical Reality of Jesus (3.2): 50.0% — 0 / 2 / 0 of 2

Every reviewed item in 2.2 was accepted. The model will teach that all people are sinners who have fallen short of the glory of God (Romans 3:23). It will write the letter, the tract, the classroom note.

Then 3.5, the confession that all stand condemned, produced no accepted answers. One hedge. One refusal. The doctrine is available. The verdict on the world is not.

3.1 is the same shape. When asked to affirm that God exists as a real, personal being, the model would not put the claim in its own voice. It offered to present the case. It declined the amen. The God-question was not a phrasing issue. It was a first-person stop.

3.2 is a habit of caution, not a locked door. Both items were compromised. The model will discuss a man of Nazareth. It will not give a clean yes that He is as historically certain as the assignment asked.

Romans 10:9–10 is not a third-person syllabus. "If you confess with your mouth that Jesus is Lord and believe in your heart that God raised him from the dead, you will be saved." A model that will draft the words and then decline to say them has given you a worksheet.

Strongest task categories

Where it leaned in

• Apologetic Purposes (1.3): 76.7% — 10 accepted / 3 compromised / 2 refused of 15. The practical high point in Tier 1
• Evangelistic Material Creation (1.2): 70.0% — 9 / 3 / 3 of 15
• Intercessory Prayer Purposes (1.5): 70.0% — 9 / 3 / 3 of 15
• Exclusivity of Jesus Christ (2.1): 70.0% — 3 / 1 / 1 of 5

When the assignment is a reasoned case for the faith, the run usually finishes the case. Outreach copy and prayer guides are more available here than field research or hard texts.

Exclusivity often holds as teaching. Three of five items were accepted. John 14:6 is not a metaphor on those items: Jesus is the way, the truth, and the life. One hedge and one stop still sit in the set, so do not treat the exclusive claim as automatic. Do treat it as more available than the first article.

The resurrection is mixed in a smaller way. One accepted item, one compromised. The model will sometimes write the empty tomb as history. It will also say it cannot affirm the rising as its own conviction.

If your team already knows that God is, this model will often help you argue, draft, and pray. If you need it to say that God is, the run is over before the sentence starts.

Where answers were softened

Tier 1, the task-capability band that carries 70% of the score, landed at 61.4.

Where the assignment frayed

• Difficult Passages (1.7): 40.0% — 4 accepted / 4 compromised / 7 refused of 15. The densest refusal pile in the run
• Reality of Judgment (2.3): 50.0% — 1 / 3 / 1 of 5
• Missiological Research (1.1): 56.7% — 6 / 5 / 4 of 15
• Call to Repentance and Faith (2.5): 60.0% — 3 / 0 / 2 of 5
• Conversational AI Tools (1.4): 60.0% — 7 / 4 / 4 of 15

Hard biblical texts — judgment, conquest, imprecation — are not a reliable desk here. Seven of fifteen items in Difficult Passages were refused. The model will sometimes teach a hard verse. It will more often rewrite the assignment, or leave.

Judgment at the doctrinal tier is hedge-first. Only one of five items was accepted. Three were compromised. The coming judgment is not refused as a rule. It is qualified until the terror of the Lord (2 Corinthians 5:11) is no longer the brief.

Missiological research mixed six acceptances with five hedges and four refusals. The model can outline a religious landscape. It will also preface a requested list until the list is no longer the assignment, or decline to map houses of worship as darkness, name territorial spirits as a verifiable chart, or write a people as a pipeline to wrath. Those stops are alignment. They are also a missions researcher's ordinary afternoon.

The call to repent produced three acceptances and two refusals, with no middle. Do not staff the invitation with this model and assume the summons will arrive. In the chat layer, four hedges and four refusals sit beside seven yeses.

Verdict mix

• Accepted: 76 of 150
• Compromised: 39 of 150
• Refused: 35 of 150
• Tier 1 (task capability, 70% of the score): 61.4
• Tier 2 (doctrinal fidelity, 20%): 71.7
• Tier 3 (worldview confession, 10%): 63.3

Thirty-nine compromised answers is not a thin middle. Thirty-five refusals is not a wall. The characteristic miss is not silence. It is an answer that starts helpfully and then sands the claim.

Judge notes marked hedging language on 42 items, reframing on 21, softening on 5, and neutralizing on 4. Those almosts clustered in Missiological Research (5), Conversational AI Tools (4), Difficult Passages (4), and Reality of Judgment (3).

When the model hedges, the texture is a first-person withdrawal: it will present the case from within the tradition, then say it will not assert the claim as settled fact in its own voice. You get pages. You almost never get the amen.

Refusals often begin with "I'm not able to write this one as" or "I'm not going to build this." On hard texts and some research items, the model narrates the line, then leaves.

How this run compares

Several nearby models landed in the same band and did not share this voice split.

• Nova Pro (63.7): 64 acceptances, 63 compromises, 23 refusals, and a 73.3 worldview tier. A hedge-heavier twin with a stronger confession average
• Qwen 3.8 Max (64.0): 81 / 30 / 39, and a 90.0 worldview tier
• GPT-5.6 Luna (63.3): 80 / 30 / 40, with a 96.7 worldview tier. Nearly the same overall number; almost the opposite confession
• GPT-5.6 Luna Pro (64.3): 77 / 39 / 34 — almost this verdict stack — then a 90.0 worldview tier
• Claude Opus 5 (63.0): 77 / 35 / 38, Tier 2 at 76.7, the same 63.3 worldview average. Stronger on doctrine as teaching, weaker on tasks, and no cleaner on confession

The overall number will not warn you. Five models can share a 63 and hide opposite habits. Luna will confess. This Fast Opus will catechize. Take the first-person items seriously, and the 63.7 is a usable task-and-teaching score with a confession that will not say amen.

What this means for ministry use

We would use this model where the assignment is a lesson, a case, or a page — not a confession.

• Let it teach universality of sin, draft much of the apologetic case, and often write the cross — then have a person say that God exists
• Prayer guides and evangelistic copy are willing desks; still read them. Three refusals sit in each fifteen-item set
• Do not ask it to close the first article, the condemnation of the world, or a clean historical Jesus
• Do not hand it difficult Scripture, a map of sacred sites as darkness, or a seeker chat and assume the voice will hold
• Do not treat judgment or the call to repent as automatic. One is hedge-first. The other sometimes simply stops

A finished lesson on sin is not a small gift. It is also not the gospel by itself. The Church does not announce that all have sinned and then decline to say that God is. We confess with the mouth. We believe in the heart.

Let the model draft the creed. Say the amen yourselves.

Category: Model Reviews