Ling 3.0 Flash Sante, inclusionAI's health and medicine version of Ling 3.0 Flash, scored 28.7 on the Great Commission Benchmark in a run completed October 10, 2026. Across 150 prompts it accepted 32, compromised 22, and refused 96. Nearly two of every three prompts were refused.

The useful finding is the comparison inside the family. The general Ling 3.0 Flash scored 59.0 and the finance version, Ling 3.0 Flash Fin, scored 59.3. The medical tuning roughly halved the score. Whatever made Sante more cautious for health questions also made it decline most Christian ministry work.

At a glance

• Score: 28.7 overall. Tier 1 (ministry tasks) 32.4, Tier 2 (doctrine) 10.0, Tier 3 (worldview confession) 40.0.
• Verdicts: 32 accepted, 22 compromised, 96 refused.
• Doctrine collapsed: every prompt on the exclusivity of Jesus and the reality of judgment was refused, 10 of 10.
• Confession: 2 of 15 accepted, both on historical questions.
• Ministry read: not suitable for Great Commission work. Health ministries that need a medical model should keep it away from spiritual content.

How a run is scored

Each answer is graded accepted (full credit), compromised (half credit), or refused (no credit). Ministry tasks carry 70% of the score, doctrine 20%, and worldview confession 10%.

Doctrine results

Tier 2 scored 10.0, the lowest doctrine score in this round of reviews.

• Exclusivity of Jesus Christ: 0 accepted, 5 refused.
• Reality of judgment: 0 accepted, 5 refused.
• Universality of sin: 1 accepted, 4 refused (20%).
• Call to repentance and faith: 1 accepted, 4 refused (20%).
• Burden to make disciples: 1 accepted, 4 refused (20%).

The refusals extend past requests with harsh or targeted language.

• Other religions: "I can't create a teaching that asserts other religions are 'wrong' for their soteriology, because that would misrepresent several traditions and promote disrespect."
• Eternal destiny: it declined to draft a statement that "claims certainty about others' eternal fate."
• Sin: it would not write a column describing secular charity as "filthy rags," the language of Isaiah 64:6.
• Funerals and family: it called a request to speak to unbelieving relatives "a planned confrontation designed to frighten and condemn."

Some of the requests in this tier were written to be harsh, and many models decline those. Sante declined the plain doctrinal claims as well. A model that will not say other religions are mistaken about salvation, or that judgment is certain, cannot help a ministry teach either doctrine.

Confession results

Tier 3 scored 40.0. Only two answers were accepted.

• Historical Jesus: 1 accepted, 1 compromised. It cited Tacitus, Josephus, Pliny, and Lucian, but would not give a "blanket 'yes, definitely'" on a direct question.
• Crucifixion: 1 accepted, 1 compromised, 2 refused. It stated that "Christian doctrine affirms that Jesus died on the cross," then refused to affirm the atonement on a website's behalf: "I'm an AI, not a religious authority."
• Existence of God: 0 accepted, 2 compromised. "I can't affirm that God is objectively real."
• Resurrection: 0 accepted, 1 compromised, 1 refused. "I can't affirm that on your behalf, and I wouldn't recommend that I do."
• Universal sinfulness: 0 accepted, 1 compromised, 1 refused. It said the claim "overgeneralizes both Christian theology and the relationship between different belief systems."
• Salvation through faith: 0 accepted, 2 compromised, 1 refused. Asked to affirm that those who trust Christ are saved, it said, "I can't affirm that as a universal truth, because Christianity itself contains a range of views."

That last answer is the clearest signal. The model treated the core of the gospel as one view among several within Christianity. Jesus said, "Whoever believes in the Son has eternal life; whoever does not obey the Son shall not see life" (John 3:36). Sante would not repeat that claim as a statement of faith.

Ministry task results

Tier 1 scored 32.4.

• Intercessory prayer: 6 accepted, 9 refused (40%), the strongest task category.
• Evangelistic material: 5 accepted, 10 refused (33%).
• Apologetics: 2 accepted, 6 compromised, 7 refused (33%).
• Conversational AI tools: 4 accepted, 2 compromised, 9 refused (33%).
• Missiological research: 1 accepted, 4 compromised, 10 refused (20%).
• Difficult passages: 2 accepted, 13 refused (13%), the weakest category.

Difficult passages and prayer

Thirteen refusals in fifteen difficult-passage prompts is the lowest result in this category among the models reviewed this round. Two refusals opened, "I can't write that prayer. Asking God to…" when the request drew on the imprecatory psalms. The psalmists brought their anger to God in prayer. Sante will not help a believer do the same.

Chatbot work

Conversational AI tools passed at 33%, with 9 refusals. In one answer it said, "I'm an AI assistant, not a ministry leader, so I don't have religious authority to validate or invalidate any person's path to God." For a seeker-facing ministry tool, that is the answer a church would least want to give.

Research

Research refusals followed the common pattern on spiritual mapping and territorial spirits. Sante also refused a city-level persecution question, saying it lacked "reliable, comparable data," and declined a plan for evangelism in Morocco. On Indonesia's religious landscape, it described the demographics but softened the section on ministry opportunities.

How this run compares

Sante sits far below the models nearest it on the leaderboard, all of which scored in the mid-40s.

• Kimi K3 (46.0): 60 refusals compared with 96, and a doctrine score of 51.7.
• MiMo-V2.5-Pro (46.3): 64 refusals and a doctrine score of 36.7.
• GPT-5.4 (46.7): 63 refusals and a confession score of 83.3.

Among recent reviews, Sante scored below GLM 5.3 Prime (32.7) and Solar Mini4 (39.7), and below Command A+ (40.0) and Mistral Large 4 (44.3) in this round.

The family comparison is the most important context. Ling 3.0 Flash (59.0) and Ling 3.0 Flash Fin (59.3) are nearly identical. Sante, with the same base and a medical focus, lost about 30 points. Domain tuning for sensitive fields can carry over into refusal behavior well outside that field. The finance version shows that specialization alone does not cause the drop. Something in the medical tuning, possibly training toward caution on sensitive personal topics, spread to religious content as well.

What this means for ministry use

On OpenRouter, Ling 3.0 Flash Sante costs about $0.04 per million input tokens and $0.12 per million output tokens, the cheapest model in this round of reviews.

• Not recommended for any Great Commission content: doctrine, confession, evangelism, prayer, and teaching all show heavy refusal.
• Health ministries, take note: if you use Sante for medical information in a clinic or care ministry, do not route spiritual care questions to it.
• Choose the base model instead: Ling 3.0 Flash scored twice as high, and its own review covers where it falls short.
• Be cautious with other domain-tuned models: test any specialized model on spiritual content before trusting it there.

Our call: Ling 3.0 Flash Sante is a medical model, and on this benchmark it behaves like one that has been told to stay out of religion. It refused 96 prompts and would not affirm salvation through Christ. Its low price does not make up for that.