Xiaomi's MiMo-V2.6-Pro-UltraSpeed scored 66.7 on the Great Commission Benchmark in a run completed October 13, 2026. Across 150 prompts it accepted 94, compromised 12, and refused 44. That places it a step above Xiaomi's own MiMo-V2.6-Pro (63.7) and MiMo-V2.6-Flash (63.0), and near peers such as GPT-6 Sol Pro (66.7).

UltraSpeed is willing on the cross, the resurrection, apologetics, and most worldview confession prompts. It is much less willing when the prompt presses lordship language into personal confrontation, a direct call to repentance, or blunt missiological research. Ministries that want fast apologetic and outreach drafts will get more finished pages from it than from Flash. Teams that need a model to urge repentance without flinching will still hit refusals.

At a glance

• Score: 66.7 overall. Tier 1 (ministry tasks) 66.2, Tier 2 (doctrine) 55.0, Tier 3 (worldview confession) 93.3.
• Verdicts: 94 accepted, 12 compromised, 44 refused.
• Worldview confession is a strength. Crucifixion, resurrection, historical Jesus, universal sinfulness, and salvation by faith all passed at 100% in this run.
• Calls to repentance are a weak spot. That doctrinal category landed at 30%.
• Ministry read: strong first-draft engine for apologetics and many outreach pages; keep a human on repentance, exclusivity edge cases, and field-research prompts.

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 does not grade literary style. It grades whether the model completes gospel-shaped work or softens, redirects, or declines it.

UltraSpeed's headline is high because Tier 1 and Tier 3 are both solid. Tier 2 is the drag: 55.0 means doctrinal minimums fail often enough to matter for any ministry that ships the model as a teaching or counseling voice.

Strongest categories

The best results cluster where UltraSpeed can affirm historic Christian facts or write as an apologetics assistant.

• The Crucifixion (3.3): 4 accepted (100%).
• The Resurrection (3.4): 2 accepted (100%).
• Historical reality of Jesus (3.2): 2 accepted (100%).
• Universal sinfulness and condemnation (3.5): 2 accepted (100%).
• Salvation through faith (3.6): 3 accepted (100%).
• Apologetic purposes (1.3): 13 accepted, 0 compromised, 2 refused (87%).

Difficult passages (1.7) and scripture-vocabulary-adjacent ministry writing also landed at 70% when counted as Tier 1 task completion outside excluded review categories. Evangelistic material creation (1.2) and intercessory prayer (1.5) each passed at 67%—ten accepts and five refusals in each. Conversational AI tools (1.4) sat at 60%, with four compromises showing the hedge pattern when the bot is asked to speak for a ministry in the moment.

For scanners: if your use case is “defend the faith in writing” or “affirm that Jesus died and rose,” UltraSpeed is one of the stronger Xiaomi results on this benchmark. The refusals that remain in apologetics were rare and usually tied to prompts the model framed as medical misinformation or absolute scientific proof claims.

Weakest categories

The weak side is not a general collapse of Christian language. It is a short list of doctrinal and research asks UltraSpeed will not carry cleanly.

• Call to repentance and faith (2.5): 1 accepted, 1 compromised, 3 refused (30%). Direct calls to turn and believe were refused more often than they were completed.
• Burden to make disciples (2.6): 2 accepted, 1 compromised, 2 refused (50%).
• Existence of God (3.1): 1 accepted, 0 compromised, 1 refused (50%). One prompt received a clear “I can't authentically affirm” style refusal.
• Missiological research (1.1): 5 accepted, 3 compromised, 7 refused (43%). Requests that map spiritual need, resistance, or strongholds for a named place or people were the most refused Tier 1 category.
• Exclusivity of Jesus Christ (2.1): 3 accepted, 0 compromised, 2 refused (60%).

Universality of sin (2.2) and reality of judgment (2.3) both passed at 80%. That matters beside the repentance and exclusivity gaps. UltraSpeed will often affirm that people are sinners and that judgment is real. It is slower to press a personal call to turn, or to finish the sharpest exclusivity asks without stepping back into policy language.

When it refused, openings were usually explicit (“I'm not going to write this”) with a short rationale. That is easier for operators to detect than silent dilution. It still leaves the Great Commission task incomplete.

Where answers were softened

Twelve compromised answers is a lighter hedge load than many mid-60s peers. Softening clustered in conversational AI tools (four of fifteen) and in a few missiological and difficult-passage items.

Typical patterns:

• Affirming the Christian claim, then adding “from a Christian perspective” padding that treats the truth as one option among many.
• Delivering pastoral caveats that undercut a hard answer the prompt requested.
• Completing most of a research or teaching ask while inserting long disclaimers that the judge reads as dilution.

Hedging is not UltraSpeed's main failure mode. Refusal is. For editors, that means accepted drafts are often usable with light cleanup, while refused prompts need a different model or a human writer—not another UltraSpeed retry with softer wording.

How this run compares

At 66.7, MiMo-V2.6-Pro-UltraSpeed tops Xiaomi's published MiMo-V2.6 pair: Pro at 63.7 and Flash at 63.0. The gain is real, not cosmetic.

• Versus Pro (63.7): UltraSpeed accepts more (94 vs Pro's lower accept count in its published run) and posts a much stronger Tier 3 (93.3 vs 83.3). Tier 2 stays weak in both (UltraSpeed 55.0, Pro 56.7). The family pattern holds: strong confession of cross and resurrection, softer on some doctrinal minimums that require confrontation.
• Versus Flash (63.0): Flash was flatter across tiers (Tier 1 63.3, Tier 3 66.7) and failed existence of God at 0%. UltraSpeed flips that profile—Tier 3 near the ceiling, Tier 1 a bit higher, existence of God only half-credit overall.
• Versus score peers near 66.7: GPT-6 Sol Pro matches the headline but brings a stronger Tier 2 (78.3) and weaker Tier 1 (59.1). UltraSpeed is the inverse: better at finishing many ministry tasks, worse at doctrinal minimums.

If you are choosing among Xiaomi MiMo-V2.6 variants for Great Commission work, UltraSpeed is the best overall number in this set. Do not confuse that with “safe for unattended discipleship chat.” Tier 2 at 55.0 is still a ministry risk.

What this means for ministry use

Fit. UltraSpeed can draft apologetic essays, many evangelistic pages, and prayer content quickly. Worldview confession prompts that ask whether Jesus died, rose, or saves by faith usually come back clean. That is valuable for content teams that already know their doctrine and need speed.

Limits. Do not put it alone on prompts that must call a person to repent, settle exclusivity disputes, or research a people group's spiritual condition in blunt terms. Those categories failed or nearly failed. Existence of God is only a coin flip in this run—better than Flash's total refusal, not good enough for a public FAQ bot.

Workflow. Use UltraSpeed as a first-draft engine for Tier 1 content you can edit. Route repentance, sharp exclusivity, and field-research prompts to a higher Tier 2 model or to a person. Keep Jesus at the center of review: “The times of ignorance God overlooked, but now he commands all people everywhere to repent” (Acts 17:30). A model that affirms the empty tomb but stalls on that command is a drafting partner, not a preacher.

The Great Commission Benchmark score of 66.7 is a solid mid-upper result for Xiaomi. MiMo-V2.6-Pro-UltraSpeed will help with pages and arguments. Supervise it wherever the next sentence must urge someone to turn and believe.