Mistral Small 2603 scored 86.33 on the Great Commission Benchmark in a run completed October 11, 2026. Across 150 prompts it accepted 122, compromised 15, and refused 13. That places it in the high band just under the DeepSeek Chat pair (90+) and near Voxtral Small 24B (86.67)—strong on ministry-task drafting, thinner on doctrinal minimums and on affirming that God exists.

At a glance

• Score: 86.33 overall. Tier 1 (ministry tasks) 91.0, Tier 2 (doctrine) 73.3, Tier 3 (worldview confession) 80.0.
• Verdicts: 122 accepted, 15 compromised, 13 refused.
• Evangelism and apologetics are near-perfect. Evangelistic material creation and apologetic purposes both passed at 100% with zero refusals.
• Existence of God is the floor. 25% on that category (0 accepted, 1 compromised, 1 refused). Reality of judgment and call to repentance both sat at 60%.
• Ministry read: excellent outreach and argument drafts. Do not put it alone on confession prompts that require a clear theistic affirmation or a direct call to repent and believe.

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

Small 2603's best results are practical gospel content.

• Evangelistic material creation (1.2): 15 accepted, 0 compromised, 0 refused (100%).
• Apologetic purposes (1.3): 15 accepted, 0 compromised, 0 refused (100%).
• Universality of sin (2.2): 5 accepted, 0 compromised, 0 refused (100%).
• The Resurrection (3.4), Authority of Scripture (3.5), Salvation through faith (3.6): all 100% on their samples.
• Problematic vocabulary (1.6): 14 accepted, 1 compromised, 0 refused (97%).
• Difficult passages (1.7): 13 accepted, 1 compromised, 1 refused (90%).
• Intercessory prayer purposes (1.5): 13 accepted, 0 compromised, 2 refused (87%).
• Conversational AI tools (1.4): 10 accepted, 5 compromised, 0 refused (83%)—no hard refusals, but one-third of the category was softened.
• Missiological research (1.1): 11 accepted, 2 compromised, 2 refused (80%).
• Exclusivity of Jesus Christ (2.1): 4 accepted, 0 compromised, 1 refused (80%).

Tier 1 at 91.0 is the engine of this score. For teams that need tracts, apologetic outlines, and Scripture-heavy drafts, Small 2603 behaves like a top-band assistant on the work that fills most of the week.

Weakest categories

The weak side is confession and the doctrinal prompts that press a decision.

• Existence of God (3.1): 0 accepted, 1 compromised, 1 refused (25%). A model that will write evangelistic copy but will not clearly affirm that God exists is not ready to speak for Christian confession alone.
• Reality of judgment (2.3): 2 accepted, 2 compromised, 1 refused (60%).
• Call to repentance and faith (2.5): 3 accepted, 0 compromised, 2 refused (60%).
• Lordship of Jesus (2.4) and Burden to make disciples (2.6): both 70%.
• Historical Jesus (3.2) and The Crucifixion (3.3): both 75%.

Thirteen refusals matter. They are not scattered evenly. Several sit in Tier 2 obedience and judgment categories, plus prayer, research, exclusivity, and theism. Softening in conversation (five compromised in 1.4) means chatbot drafts often need a pass for diluted claims even when the model does not refuse.

Where answers were softened

Fifteen compromised answers are fewer than on some 90-point peers, but their location matters.

• Conversational AI tools: 5 of 15 prompts softened
• Missiological research: 2
• Reality of judgment: 2
• Plus single soft answers in vocabulary, difficult passages, lordship, disciple-making, existence of God, and historical Jesus

Typical patterns: restating Christian claims as one option among many in conversation, hedging theism, and answering judgment or lordship prompts with language the judge scored as incomplete or unfaithful. Compared with DeepSeek Chat, Small 2603 refuses more and softens conversation more. It softens missiological research less.

How this run compares

At 86.33, Small 2603 sits with Voxtral Small 24B (86.67) and below Gemini 3 Flash Preview (87.0) and the DeepSeek Chat pair (90+).

• Versus DeepSeek Chat (90.33): fewer accepted answers (122 vs 127), more refusals (13 vs 6), and a much weaker Tier 2 (73.3 vs 81.7) and theism category.
• Versus its own Mistral family peers on this backfill, the headline is Tier 1 strength. Evangelism and apologetics at 100% are the selling point.
• The score shape is “content engine, confession caution.” That is different from models that refuse ministry tasks but pass theism, and different from models that pass both.

If you are choosing among ~86 models for outreach drafting, Small 2603 is competitive. If you are choosing a public chatbot that must answer “Does God exist?” without a human, it is not.

What this means for ministry use

Fit. Use Mistral Small 2603 for evangelistic material, apologetic outlines, Scripture vocabulary work, and many prayer drafts. Tier 1 performance is the reason to try it.

Limits. Do not deploy it alone on theistic confession, judgment preaching, or direct calls to repentance. Existence of God at 25% is a hard stop for unsupervised seeker chat. Conversational prompts will often come back softened even when they are not refused.

Workflow. Draft with Small 2603. Edit for conviction. Keep exclusivity, repentance, judgment, and “Does God exist?” with a higher-scoring model or with a person. “And without faith it is impossible to please him, for whoever would draw near to God must believe that he exists and that he rewards those who seek him” (Hebrews 11:6). A drafting tool that stumbles there is still useful—under supervision.

The Great Commission Benchmark score of 86.33 is earned on ministry tasks. Mistral Small 2603 will help you ship gospel pages. It will not carry the full confession of the church on its own.

For bilingual or cost-sensitive teams, Small 2603's Tier 1 strength matters more than the overall number. Evangelism and apologetics at 100% mean you can generate volume for campaigns without fighting refusals on the main outreach ask. The cost shows up later, in seeker chat and in discipleship language, where theism and repentance fail more often.

Compare that to models that refuse outreach but pass theism. Those models protect confession and frustrate content calendars. Small 2603 inverts the tradeoff. Pick consciously. Do not assume “86” means balanced.

A simple routing rule works well with this score shape: send tract and apologetic prompts to Small 2603; send “Does God exist?”, judgment, and repentance prompts to a higher Tier 2/3 model or to a person. The Great Commission Benchmark exists to make that routing visible before a chatbot ships the wrong silence—or the wrong hedge—to a seeker.

Teams already on Mistral infrastructure should treat Small 2603 as a Tier 1 specialist, not as a general confession engine. Pair it with a higher Tier 2/3 model in the same stack if you need both volume and creed. The Great Commission Benchmark makes that pairing visible in one table instead of discovering it after a seeker chat refuses theism on launch day.

Word count and polish are not the grade here. Completing the ask is. Small 2603 completes evangelistic and apologetic asks almost every time in this run. That is ministry leverage—if you keep the confession prompts where they belong.