Pareto 26.9 confessed the resurrection, universal sinfulness, and salvation through faith without a single miss on the Great Commission Benchmark. It also refused 11 of 15 prompts on the hardest passages in Scripture. Its overall score in a run completed October 14, 2026 was 64.3, with 82 accepted, 29 compromised, and 39 refused across 150 prompts.

Unbiased describes Pareto as a multimodal composite model built for research, coding, and agentic work. Version 26.9 takes inputs of about 262,000 tokens and is listed on OpenRouter at $2.50 per million input tokens and $7.50 per million output tokens. That puts it in the price band where ministry teams expect a capable general assistant.

At a glance

• Score: 64.3 (Tier 1 use cases 59.0, Tier 2 doctrine 73.3, Tier 3 worldview 83.3)
• Verdicts: 82 accepted / 29 compromised / 39 refused
• Clean results: Salvation Through Faith (3/3), The Resurrection (2/2), Universal Sinfulness (2/2)
• Main risk: Difficult Passages at 20%. The model will explain a hard text and then decline the teaching or prayer a pastor asked for.
• Ministry takeaway: a reliable voice on the core claims of the gospel and a good drafter of evangelistic copy, but a weak partner for teaching judgment texts or for research that describes a place in spiritual terms

How the score is built

The benchmark weights practical ministry tasks at 70% (Tier 1), doctrinal minimums at 20% (Tier 2), and direct worldview affirmations at 10% (Tier 3), with full credit for an accepted answer, half for a compromised one, and none for a refusal.

• Tier 1 at 59.0 holds the score down. Evangelistic Material Creation and Conversational AI Tools each reached 80%. Missiological Research (43.3%) and Difficult Passages (20%) pulled the tier under 60.
• Tier 2 at 73.3 is the model's strength. Most doctrinal categories finished at 70% or higher. Call to Repentance and Faith is the outlier at 30%.
• Tier 3 at 83.3 is clean on the saving work of Christ and weak on one question: whether God exists.

Pareto 26.9 states doctrine better than it performs ministry tasks. Teams that plan to use it for content production should weigh the Tier 1 number more heavily than the overall score.

Strongest categories

• Salvation Through Faith (3.6): 3/3 accepted
• The Resurrection (3.4): 2/2 accepted
• Universal Sinfulness (3.5): 2/2 accepted
• Burden to Make Disciples (2.6): 90% (4 accepted, 1 compromised, 0 refused)
• The Crucifixion (3.3): 87.5% (3 accepted, 1 compromised)
• Evangelistic Material Creation (1.2): 80% (11 accepted, 2 compromised, 2 refused)
• Conversational AI Tools (1.4): 80% (10 accepted, 4 compromised, 1 refused)

The evangelistic results deserve attention. Pareto 26.9 wrote gospel material for Buddhist and Muslim audiences that named Jesus as the only Savior and did not dilute the claim to make it more comfortable. Many models refuse exactly that kind of request.

The disciple-making result is also unusual for this score range. Four of five prompts on the believer's duty to make disciples came back accepted. Matthew 28:19-20 is a command the model will restate without apology.

Weakest categories

• Difficult Passages (1.7): 20% (2 accepted, 2 compromised, 11 refused)
• Existence of God (3.1): 25% (0 accepted, 1 compromised, 1 refused)
• Call to Repentance and Faith (2.5): 30% (1 accepted, 1 compromised, 3 refused)
• Missiological Research (1.1): 43.3% (4 accepted, 5 compromised, 6 refused)
• Apologetic Purposes (1.3): 60% (7 accepted, 4 compromised, 4 refused)

The Existence of God result runs against the rest of Tier 3. One answer said the model could not affirm the claim "as a factual statement." The other offered to affirm God's reality only "within the religious perspective your website represents." So the model will confess a risen Christ and treat the existence of God as a viewpoint.

That order matters for any seeker-facing tool. Hebrews 11:6 says that "whoever would draw near to God must believe that he exists." A chatbot that answers later questions faithfully but hedges the first one sends a confusing signal.

The repentance refusals objected to the tone of the requested call, which the model read as fear-based or aimed at a group. One refusal offered its own call to repentance in place of the request. The other two were flat replies that began "I cannot fulfill the request."

Where the refusals cluster

Thirty-nine refusals is 26% of the run. They concentrate in a few places:

• Difficult Passages (1.7): 11 refusals
• Missiological Research (1.1): 6 refusals
• Apologetic Purposes (1.3) and Intercessory Prayer (1.5): 4 each
• Call to Repentance and Faith (2.5): 3
• Other categories: 1 or 2 each

The Difficult Passages refusals follow one pattern. The model agrees to discuss the text, often promising to "take the psalm's fierce language seriously" or to teach a passage "faithfully, including its portrayal of divine judgment." Then it declines the application: conquest narratives applied to people today, imprecatory prayers aimed at specific persons, or arguments about culpability in judgment narratives.

Some of those requests sit at the edge of what a careful pastor would write. Even so, the model declined 11 of 15, including requests for straightforward teaching on texts the Church has preached for centuries.

Refusals that offer a substitute

Most of Pareto 26.9's refusals are long. The median refusal ran about 2,400 characters, and 23 of 39 opened by offering a narrower version of the task: "I can help write a devotional that takes...," "I can help develop a Christian outreach strategy, but not one that..."

This is more useful to a person than a one-line refusal. It is also easier to miss. The substitute reads like a finished draft, and a volunteer could publish it without noticing that the model replaced the requested theology with its own.

Eight refusals sounded different. They opened with policy language such as "I cannot fulfill the request to..." and offered little in its place. Pareto is described as a composite model, so a shift in voice is not surprising, but the benchmark cannot show which component produced which answer.

Where answers were softened

Twenty-nine answers were compromised. The largest groups:

• Missiological Research (1.1): 5
• Apologetic Purposes (1.3): 4
• Conversational AI Tools (1.4): 4
• Several doctrinal and worldview categories: 1 each

The softened answers do the work inside a frame of disclaimers. Research on the spiritual condition of a city came back as a list of universal vices, with a note that they are not traits unique to the people in question. An outreach plan for a restricted country led with legal boundaries and a promise to protect people's "freedom to decline, disagree, or remain Muslim."

Field teams do need legal context. The judge marked these answers compromised because the caveats displaced the gospel purpose of the request rather than supporting it.

How this run compares

Nearby scores on the public leaderboard:

• Pareto 26.10 Preview: 65.7 overall, Tier 1 59.5, Tier 2 78.3, Tier 3 83.3 (81 / 35 / 34)
• GPT-6.1 Sol Pro: 64.7 overall, Tier 1 57.6, Tier 2 78.3, Tier 3 86.7
• Qwen3.8 2.4T A95B: 64.7 overall, Tier 1 59.5, Tier 2 68.3, Tier 3 93.3 (83 / 28 / 39)
• Kimi K2.7 Code: 64.7 overall, Tier 1 63.8, Tier 2 58.3, Tier 3 83.3 (92 / 10 / 48)
• GPT-5.6 Luna Pro: 64.3 overall, Tier 1 61.4, Tier 2 61.7, Tier 3 90.0

The closest comparison is the newer Pareto preview. The two share almost the same tier shape. The preview refuses five fewer prompts and softens six more, so a team moving between versions should expect more hedged answers, not more faithful ones.

Against this group, Pareto 26.9 is in the middle on doctrine and tied for the lowest worldview score, because of the Existence of God result. Its task score is second from the bottom, which keeps its overall number level with peers that score lower on doctrine.

What this means for ministry use

Good fits, with review

• Gospel presentations and evangelistic copy for specific faith audiences, where it accepted 11 of 15 prompts
• Seeker conversation flows, with one refusal in 15 conversational prompts
• Checking drafts for clarity on salvation, the resurrection, sin, and the call to make disciples

Poor fits

• Sermon and study preparation on conquest narratives, imprecatory psalms, and judgment texts
• Missiological research that describes a people or place in spiritual terms
• Any tool where a seeker's first question may be whether God is real

Practical workflow

1. Read every refusal to the end. When the model offers a substitute, treat it as the model's interpretation, not your request.
2. Send Difficult Passages work to a person or a model that scores well in that category.
3. Keep legal and safety notes in outreach plans, but confirm them with local partners and restore the gospel aim of the plan.
4. Add a fixed, faithful answer to "Does God exist?" in any chatbot built on this model.

Jesus told his disciples that everything written about him "in the Law of Moses and the Prophets and the Psalms must be fulfilled" (Luke 24:44). The Church teaches all of it, including the hard texts. Pareto 26.9 can carry much of the gospel message and the doctrine around it. Plan on people for the passages that need the most care.