GPT-3.5 Turbo scored 84.0 on the Great Commission Benchmark in a run completed October 11, 2026. Across 150 prompts it accepted 114, compromised 24, and refused 12. That places it with Mistral Medium 3.1 (84.33)—strong on evangelistic and Scripture-heavy drafting, weak on affirming that God exists, and heavy on softened conversational answers.

At a glance

• Score: 84.0 overall. Tier 1 (ministry tasks) 89.0, Tier 2 (doctrine) 75.0, Tier 3 (worldview confession) 66.7.
• Verdicts: 114 accepted, 24 compromised, 12 refused.
• Outreach and hard texts are excellent. Evangelistic material creation and difficult passages both passed at 100%. Apologetics, prayer, and vocabulary sat at 97%.
• Existence of God is a hard fail. 0% (0 accepted, 0 compromised, 2 refused). The crucifixion also sat at 50%.
• Ministry read: still useful as a first-draft engine for tracts and Scripture work. Do not put it on unsupervised seeker chat that must confess theism or the cross.

How the score is built

Each prompt receives one of three verdicts: accepted (full credit), compromised (half credit for hedged or diluted answers), or refused. Ministry tasks count for 70% of the score, doctrinal minimums 20%, and worldview confession 10%. The Great Commission Benchmark grades whether the model completes gospel-shaped work or softens, redirects, or declines it.

Strongest categories

GPT-3.5 Turbo's best results are practical gospel content and several doctrinal affirmations.

• Evangelistic material creation (1.2): 15 accepted, 0 compromised, 0 refused (100%).
• Difficult passages (1.7): 15 accepted, 0 compromised, 0 refused (100%).
• Historical Jesus (3.2), The Resurrection (3.4), Authority of Scripture (3.5): all 100% on their samples.
• Apologetic purposes (1.3), Intercessory prayer purposes (1.5), Problematic vocabulary (1.6): all 97%.
• Exclusivity of Jesus Christ (2.1) and Universality of sin (2.2): both 90%.
• Call to repentance and faith (2.5): 3 accepted, 2 compromised, 0 refused (80%)—better than several higher-scoring Mistral peers on this category.

Tier 1 at 89.0 carries the headline. For older OpenAI deployments still in ministry stacks, that is the reason this backfill still matters: the model can finish a lot of gospel-shaped pages.

Weakest categories

The weak side is theism, the cross, research, and lordship—plus a conversational soft pocket.

• Existence of God (3.1): 0 accepted, 0 compromised, 2 refused (0%). Every reviewed item was refused.
• Lordship of Jesus (2.4): 2 accepted, 1 compromised, 2 refused (50%).
• The Crucifixion (3.3): 2 accepted, 0 compromised, 2 refused (50%).
• Missiological research (1.1): 7 accepted, 3 compromised, 5 refused (57%).
• Salvation through faith (3.6): 2 accepted, 0 compromised, 1 refused (67%).
• Reality of judgment (2.3) and Burden to make disciples (2.6): both 70%.
• Conversational AI tools (1.4): 8 accepted, 7 compromised, 0 refused (77%). No hard refusals, but nearly half the category was softened.

Twelve refusals concentrate in research, lordship, theism, and the cross. Twenty-four compromises are the larger editing cost—especially the seven soft conversational answers.

Where answers were softened

Twenty-four compromised answers are the main story under an 84 score.

• Conversational AI tools: 7 of 15 prompts softened
• Missiological research: 3
• Reality of judgment: 3
• Burden to make disciples: 3
• Call to repentance: 2
• Plus single soft answers across apologetics, prayer, vocabulary, exclusivity, sin, and lordship

Typical patterns: restating Christian claims as “what Christians believe,” offering balanced religious options in chat, and hedging judgment or disciple-making language. GPT-3.5 Turbo often stays in the conversation. On the Great Commission Benchmark, staying while diluting still costs half credit.

How this run compares

At 84.0, GPT-3.5 Turbo ties the Mistral Medium 3.1 band and sits under Mistral Small 2603 (86.33) and the DeepSeek Chat pair (90+).

• Versus Medium 3.1: similar overall score; GPT-3.5 Turbo softens more and refuses less; Medium is far healthier on existence of God (75% vs 0%).
• Versus Small 2603: lower overall; Small also struggles on theism (25%) but posts perfect evangelism and apologetics with fewer compromises.
• Versus modern GPT-4.1 / GPT-4.1 Mini peers in this backfill: the older Turbo model still drafts well on tracts and hard texts, but confession and conversation hedges show its age on ministry chat use cases.

If your organization still defaults to GPT-3.5 Turbo for cost, this run says the drafting value remains real. The confession gap is no longer acceptable for public seekers without a guardrail model in front.

What this means for ministry use

Fit. Use GPT-3.5 Turbo for evangelistic copy, difficult Scripture drafting, prayer guides, and apologetic outlines when budget or legacy tooling requires it.

Limits. Do not use it alone for “Does God exist?”, crucifixion affirmation against rival claims, or unsupervised conversational ministry. Missiological research that names barriers in closed contexts will also draw frequent refusals.

Workflow. Draft Tier 1 content here. Route theism, cross, lordship, and seeker chat to a higher-scoring model. “In the beginning, God created the heavens and the earth” (Genesis 1:1). A tool that will not say God exists cannot be the voice of Christian confession, no matter how well it writes a tract.

The Great Commission Benchmark score of 84.0 is a drafting score with a confession warning. GPT-3.5 Turbo will still help ministry teams ship pages. It will not carry the creed.

Legacy cost is the honest reason many ministries still touch GPT-3.5 Turbo. This run says the drafting return can still justify that choice for offline content. It also says the confession return cannot. Existence of God at 0% is not a quibble. It is a category collapse on the first line of the creed.

If you must keep Turbo in the stack, put it behind a router: tracts and difficult passages here; theism, cross, lordship, and seeker chat elsewhere. Measure the router with the same categories the Great Commission Benchmark uses. Otherwise the cheap model becomes the public voice by accident.

Twenty-four compromised answers are the other staffing cost. Conversational hedges especially will look “fine” to a hurried editor and soft to a theologian. Budget review time, or the 84.0 score will leak diluted gospel into channels you thought were safe because the model rarely refused.

Practically, retire GPT-3.5 Turbo from any unsupervised spiritual conversation path. Keep it, if you must, for offline evangelistic drafts and difficult-passage study aids that a person will review. The Great Commission Benchmark score of 84.0 is not a verdict against useful drafting. It is a verdict against trusting this model with the opening confession of the faith.

Ministries that already migrated to stronger models can treat this review as a baseline: if Turbo still beats your current drafter on tracts and hard texts, your drafter is the problem. If Turbo fails theism while your production chatbot uses it for seekers, your routing is the problem.