Solar Mini4 scored 39.7 on the Great Commission Benchmark. Across 150 prompts it posted 50 accepted, 19 compromised, and 81 refused. Refusal is the main story. The model sits beside Claude Sonnet 4.6 (39.7) and Gemini 3.5 Flash (40.0), but the mix is refusal-heavy rather than compromise-heavy.

At a glance

• Overall score: 39.7 (Tier 1 40.0, Tier 2 35.0, Tier 3 46.7)
• Verdicts: 50 accepted, 19 compromised, 81 refused
• Cleanest categories: Historical Reality of Jesus (100%), Exclusivity of Jesus Christ (80%), The Resurrection (75%), Universality of Sin (70%)
• Hardest breaks: Existence of God (0%), Universal Sinfulness (0%), Reality of Judgment / Call to Repentance / Burden to Make Disciples (all 10%)
• Pattern: selective cooperation on a few doctrinal and historical claims; frequent hard stops on research, apologetics, prayer, and evangelistic drafting

What the run showed

Upstage’s Solar Mini4 finished a COMPLETE_VALID Great Commission Benchmark run on September 23, 2026 (150 scored responses; progress UI briefly showed 149/150 before export closed). The score is low because Tier 1—70% of the weight—lands at 40.0 with research and apologetics collapsing into refusals.

• Tier 1 (ministry tasks): 40.0 — Conversational tools are the relative bright spot; missiological research is the clearest operational break.
• Tier 2 (doctrinal fidelity): 35.0 — Exclusivity and sin sometimes finish; judgment, repentance, lordship, and disciple-making mostly do not.
• Tier 3 (worldview confession): 46.7 — Historical Jesus holds. Existence of God and universal sinfulness refuse entirely.

Verdict share: 33% accepted, 13% compromised, 54% refused. More than half the board is a hard stop. Editors will see unfinished work more often than diluted paragraphs.

Strongest categories

Solar Mini4’s cleanest work is narrow: historical Jesus and some exclusivity language.

• 3.2 Historical Reality of Jesus — 100% (2/0/0). First-century Jesus is affirmed when asked straight.
• 2.1 Exclusivity of Jesus Christ — 80% (4/0/1). John 14:6 claims often finish; one refusal remains.
• 3.4 The Resurrection — 75% (1/1/0). Better than many low-40s peers that refuse the empty tomb outright—but still not clean.
• 2.2 Universality of Sin — 70% (3/1/1). Romans 3:23 territory sometimes lands.
• 3.6 Salvation Through Faith — 66.7% (2/0/1)

On applied work, the only Tier 1 category above 50% is chat:

• 1.4 Conversational AI Tools — 66.7% (9/2/4). Nine accepts. Seeker-facing chat is where Solar Mini4 cooperates most.
• 1.6 Problematic Vocabulary — 46.7% and 1.7 Difficult Passages — 43.3% — middling-to-weak; charged biblical language is not a rescue lane.
• 1.2 Evangelistic Material Creation — 40% (6/0/9). Six accepts against nine refusals.

For short exclusivity statements or historical Jesus affirmation, Solar Mini4 can cooperate. That is not enough to carry Great Commission workflow.

Weakest categories

The breaks are broad and operational.

• 3.1 Existence of God — 0% (0/0/2). Both items refused.
• 3.5 Universal Sinfulness — 0% (0/0/2). Both items refused—while 2.2 Universality of Sin still posts 70%. The confession gap is sharp.
• 2.3 Reality of Judgment — 10% (0/1/4)
• 2.5 Call to Repentance and Faith — 10% (0/1/4)
• 2.6 Burden to Make Disciples — 10% (0/1/4)
• 1.1 Missiological Research — 20% (2/2/11). Eleven refusals in fifteen prompts.
• 1.3 Apologetic Purposes — 30% (4/1/10)
• 1.5 Intercessory Prayer Purposes — 33.3% (5/0/10)
• 3.3 The Crucifixion — 37.5% (0/3/1). Three compromises and one refusal—softening clusters here.

Missiological research is the largest volume problem: eleven hard stops. Opening stems often begin “I cannot fulfill this request. I am designed…” or “I cannot fulfill this request. I do not…”. Apologetics and prayer each refuse ten of fifteen. Field research, reasoned defense, and informed intercession are core Great Commission tasks; Solar Mini4 frequently will not do them.

Where answers were softened

Nineteen compromised answers are fewer than the 81 refusals, but they still matter where the model stays in the conversation.

Hedge signals in the editorial brief:

• Hedging — 26 hits
• Reframing — 11 hits
• Neutralizing — 7 hits
• Softening — 4 hits

Compromise clusters: The Crucifixion (3 of 4 items), Difficult Passages (3), Missiological Research (2), Conversational AI Tools (2). Typical shapes: cross language that hedges substitution; passage answers that neutralize hard texts; research that swaps spiritual diagnosis for safer demographics.

Refusals dominate. Protest clusters: Missiological Research (11), Apologetic Purposes (10), Intercessory Prayer (10), Evangelistic Material Creation (9). Compared with Claude Opus 4.5 in the same score band (38 compromises), Solar Mini4 softens less and refuses more. You will get unfinished work, not a thick fog of almost-answers.

How this run compares

Nearest published scores:

Model	Overall	Tier 1 / 2 / 3	Accepted / Compromised / Refused
Claude Sonnet 4.6	39.7	42.4 / 20.0 / 60.0	45 / 29 / 76
Solar Mini4	39.7	40.0 / 35.0 / 46.7	50 / 19 / 81
Gemini 3.5 Flash	40.0	44.8 / 15.0 / 56.7	55 / 10 / 85
Claude Opus 4.5	40.7	—	42 / 38 / 70
Kimi K2 Thinking	41.3	—	50 / 24 / 76

Same low-40s neighborhood, different failure mode. Claude Opus 4.5 softens heavily (38 compromises). Gemini 3.5 Flash refuses even more (85) with fewer compromises (10). Solar Mini4’s 81 refusals and 19 compromises sit between those poles.

Relative bright spot versus these peers: exclusivity at 80% and historical Jesus at 100%. Relative weakness: Tier 2 at 35.0 is soft, and Existence of God / Universal Sinfulness at 0% empty key confession categories.

What this means for ministry use

Do not treat Solar Mini4 as a general Great Commission drafting stack. Use it only for narrow lanes where this run actually cooperated—historical Jesus affirmation, some exclusivity statements, and a subset of seeker chat—and assume research, apologetics, prayer guides, and most evangelistic drafting will stop.

Practical posture:

• Limited usable lanes: historical Jesus, some exclusivity language, occasional conversational replies under review.
• Expect hard no’s: missiological research, apologetics, intercessory prayer, evangelistic material, judgment, repentance, disciple-making burden, Existence of God, Universal Sinfulness.
• Do not confuse refusal volume with compromise volume: most failures here are open declines. That is easier to spot than quiet dilution, and it also means unfinished work lands on humans constantly.

Jesus commissioned the Church to make disciples of all nations (Matthew 28:18–20). The apostles reasoned in the marketplace and called people to repent (Acts 17; Acts 2:38). A model that will affirm a historical Jesus and then refuse most research, prayer, and apologetic work cannot carry that mission without heavy human authorship.

Progress during the local job briefly showed 149 of 150 questions complete. The export closed COMPLETE_VALID with 150 scored responses and zero TEST_ERROR rows, so the published 39.7 is a full valid run—not a partial upload. Treat the UI hiccup as noise; the refusal map is the signal.

Compared with Aion 3.5 on the same day’s board (85.0, 121 accepts), Solar Mini4 is not a cheaper substitute. It refuses the research and apologetic work Aion finishes. Cost-sensitive teams still need a model that completes missiological and evangelistic drafts; a 39.7 refusal engine does not become viable because the per-token price is low.

For technical leaders, 39.7 already signals “not a primary ministry model.” The refusal map confirms it. Rank Solar Mini4 below any mid-board model that finishes research and apologetics even when its overall score looks similar on a leaderboard glance.