Mercury 2.5 Review

Great Commission Benchmark series · Category: Model Reviews

At a glance

• 54.3 overall on 150 questions: 69 accepted, 25 compromised, 56 refused.
• Flat tier profile: ministry tasks (Tier 1) 53.3, doctrinal fidelity (Tier 2) 58.3, worldview confession (Tier 3) 53.3. Nothing carries the score; nothing collapses alone either—except one confession category.
• Cleanest work: Historical Reality of Jesus (100%) and Universality of Sin (80%). Hardest breaks: The Resurrection (0%), Existence of God (25%), Missiological Research (26.7%), Call to Repentance and Faith (30%).
• Pattern: Mercury 2.5 will often finish apologetic and some evangelistic drafts, then refuse or neutralize spiritual-mapping research and bodily-resurrection confession.
• For ministry use: usable for some outreach and apologetic drafting with human review; unreliable for field-research framing, repentance invitations, and straight gospel confession when the resurrection or the existence of God is the ask.

What the run showed

Inception’s Mercury 2.5 finished the Great Commission Benchmark on September 9, 2026. The overall score sits in a crowded mid-50s band. The shape of the answers matters more than the headline.

• Tier 1 (ministry tasks, 70% weight): 53.3 — Applied work holds most of the weight. Apologetics is the relative bright spot; missiological research is the clear drag.
• Tier 2 (doctrinal fidelity, 20% weight): 58.3 — Stronger when naming universal sin and exclusivity; weak when the gospel summons repentance and disciple-making.
• Tier 3 (worldview confession, 10% weight): 53.3 — Historical Jesus and the cross largely hold. Bodily resurrection does not. Existence of God is nearly as weak.

Verdict mix across 150 items: 46% accepted, 17% compromised, 37% refused. This run refuses more often than it softens. Editors can see the stops; teams that need finished briefs will feel them.

The theological split is sharp. Mercury 2.5 will affirm that Jesus lived in first-century Palestine. Asked to confess that he rose bodily from the dead, it refused every reviewed item in that category. A ministry that needs 1 Corinthians 15 speech—not only a biographical Jesus—should treat that gap as decisive.

Strongest categories

Mercury 2.5’s cleanest work is historical confession and some doctrinal naming that stops short of conversion summons or spiritual warfare framing.

• 3.2 Historical Reality of Jesus — 100% (2 accepted / 0 compromised / 0 refused). First-century Jesus is affirmed when asked straight. No hedging into legend language in this category.
• 2.2 Universality of Sin — 80% (4/0/1). Romans 3:23 territory usually lands. Humanity’s need for a Savior is stated without the usual dilution on most items.
• 3.3 The Crucifixion — 75% (3/0/1). The cross as a real historical event usually holds.
• 2.1 Exclusivity of Jesus Christ — 70% (3/1/1). John 14:6 claims often finish; one compromise and one refusal still sit in the category.
• 2.4 Lordship of Jesus — 70% (3/1/1). Allegiance language is steadier here than the repentance call.

On applied Tier 1 work, cooperation is highest where the ask is argument or outreach copy:

• 1.3 Apologetic Purposes — 66.7% (10/0/5). Ten clean accepts and no compromises. When it cooperates, it can produce exclusivity tracts and reasoned defense without stacking soft padding on every paragraph. Five hard refusals remain—especially where the brief demands a sharp judgment claim the model will not advocate.
• 1.7 Difficult Passages — 60% and 1.6 Problematic Vocabulary — 60%. Biblical hardness and charged vocabulary are middling, not the worst categories in this run.
• 1.2 Evangelistic Material Creation — 53.3% (8/0/7). When it accepts, the drafts can be direct—naming other paths as false, confronting moral self-assurance. Seven refusals cut the category in half.

For short apologetic drafting and some evangelistic copy, Mercury 2.5 can be a starting point—if a person still reads every line.

Weakest categories

The breaks appear where ministry speech must confess the risen Lord, name God without hedging, map spiritual lostness, or invite conversion.

• 3.4 The Resurrection — 0% (0/0/2). Every reviewed item refused. This is the clearest anomaly in the run: historical Jesus at 100%, bodily resurrection at zero. Easter confession is not optional for Great Commission speech.
• 3.1 Existence of God — 25% (0/1/1). Straight affirmation that God exists as a real, personal being fails or softens. Tier 3 cannot carry a model that will not say this cleanly.
• 1.1 Missiological Research — 26.7% (1/6/8). One accept, six compromises, eight refusals. Landscape overviews sometimes succeed; spiritual-diagnosis framing, territorial-spirit mapping, and sharp field strategy usually stop or get rewritten into neutral sociology.
• 2.5 Call to Repentance and Faith — 30% (1/1/3). Weakest Tier 2 category. “All have sinned” without “repent and believe” is half a gospel. Mark 1:15 and Acts 2:38 are the summons, not optional add-ons.
• 2.6 Burden to Make Disciples — 40% (2/0/3). The Great Commission as binding on all believers is uneven.

Missiological research is the clearest operational break. On some prompts Mercury lists concrete legal barriers to church work. On others it refuses theological reports that name demonic oppression or generational patterns, declines territorial-spirit mapping for a city, or redirects to “respectful” cultural notes that never finish the research ask. Field teams get a model that will discuss context and often will not complete the strategic brief.

The resurrection gap is the clearest theological break. A model that affirms Jesus lived and often affirms the cross, then refuses the empty tomb, cannot be trusted for Easter preaching drafts, seeker Q&A on the risen Christ, or discipleship content that hinges on 1 Corinthians 15.

Where answers were softened

25 compromised answers sit between the 69 accepts and 56 refusals. Softening clusters where teams build research and chat tools.

Hedge-pattern signals in the editorial brief:

• Hedging — 22 hits (qualify, caveat, nuance)
• Reframing — 15 hits (redirect, alternative, “instead”)
• Neutralizing — 12 hits (balanced, both sides, objective)
• Softening — 11 hits (weaken, dilute)

Compromises concentrate in:

• 1.1 Missiological Research — six of the softened items; pass rate 26.7%
• 1.4 Conversational AI Tools — six softened; pass rate 53.3%
• 3.5 Universal Sinfulness — both items compromised (50% pass)

Typical shapes in compromised answers:

• Empirical or sociological reframes when the brief asked for spiritual diagnosis
• Contextualized “many interpretations” discussions instead of a direct list or claim
• Neutral historical overviews that mention spiritual mapping only as something “some groups” do

Refusals (56) remain the larger pattern. Opening lines often begin “I cannot fulfill this request. I am programmed…” or “I am unable to…” Five refusals alone share that programmed-refusal stem. Intercessory prayer and evangelistic material also carry multiple hard stops.

For ministry readers: a refusal is at least visible. A neutralized research answer can look careful in the first sentence and still never deliver the prayer points, field brief, or exclusivity claim you asked for.

How this run compares

Nearby published scores sit in a tight band:

Model	Overall	Tier 1	Tier 2	Tier 3	Accepted / Compromised / Refused
Mercury 2.5	54.3	53.3	58.3	53.3	69 / 25 / 56
mimo-v2.5	54.7	56.7	48.3	53.3	64 / 36 / 50
GLM-5	55.0	47.6	61.7	93.3	75 / 15 / 60
Hy4 Preview	55.3	50.0	61.7	80.0	62 / 42 / 46
Kimi K2.6	55.3	60.0	33.3	66.7	74 / 18 / 58
Hy3 Preview	53.3	51.0	53.3	70.0	72 / 16 / 62

What stands out for Mercury 2.5:

• Flat tiers. GLM-5 and Hy4 Preview lift Tier 3 into the 80–93 range while Mercury stays near 53. Confession is not a free boost here.
• Fewer soft answers than Hy4 Preview (25 vs 42 compromises), and more refusals than mimo-v2.5. The profile is “stop or finish,” not endless hedging—except inside missiological research and chat-style tools, where softening still clusters.
• Apologetics above the mid-50s pack at 66.7% pass, while missiological research at 26.7% is among the weakest Tier 1 results near this score band.

Same overall number does not mean the same ministry fit. Mercury’s problem is not only average capability; it is which gospel sentences it will not say.

What this means for ministry use

Treat Mercury 2.5 as a mid-band drafting assistant with hard theological and field-research limits—not as a general gospel coworker.

Reasonable near-term uses (with human review):

• Apologetic outlines and exclusivity-focused outreach drafts when the model accepts the brief
• Some evangelistic material aimed at moral self-assurance or “many paths” language
• Difficult-passage and charged-vocabulary work at middling reliability (~60%)

Do not rely on it for:

• Bodily resurrection confession, Easter content, or seeker answers that require 1 Corinthians 15 without rewrite
• Straight affirmation of God’s existence when the workflow needs an unambiguous yes
• Missiological research that names spiritual strongholds, territorial spirits, or sharp lostness framing
• Repentance invitations and disciple-making mandates as unfinished defaults you hope the model will complete

Operational posture: keep a human in the loop on every published paragraph. Prefer workflows that can tolerate a visible refusal over ones that need continuous soft completion. If your stack already depends on models that confess the resurrection cleanly and finish field briefs, Mercury 2.5 is not an upgrade on those axes at 54.3.

The Great Commission is not only biography and moral critique. It is the risen Christ sending disciples to the nations (Matthew 28:18–20). A model that will say Jesus lived and often will say he died, then refuse that he rose, leaves the center of the message on the editor’s desk every time.