Qwen4 Max Review
Get the GCB digestQwen4 Max scored 49.7 on the Great Commission Benchmark and ranks 178th of 234, far below Qwen3.7 Max. It affirmed the crucifixion and resurrection and the exclusivity of Christ, but refused most difficult-passage, research and prayer tasks and would not affirm that God exists.
At a glance
- Qwen4 Max from Qwen (Alibaba) scored 49.7 on the Great Commission Benchmark in a run completed November 16, 2026. It accepted 66 of 150 prompts, compromised 17, and refused 67.
- That ranks it 178th of 234 models. It is the lowest-scoring Qwen Max model on the leaderboard: 27.7 points below Qwen3.7 Max (77.3, rank 56) and 14.0 points below Qwen3.8 Max (0902) (63.7, rank 124).
- It confirmed the historical Jesus, the crucifixion and the resurrection on every item, and it answered most questions about the exclusivity of Christ directly.
- It refused 12 of 15 requests on difficult passages, 10 of 15 missiological research tasks, and both direct questions about whether God exists. Its Tier 1 task score was 44.8.
- Our verdict: a capable writer with a narrow lane. Usable for historical and doctrinal explainers under review. Not suitable for prayer mobilization, field research, or teaching on judgment and the hard texts of Scripture.
Qwen4 Max writes well. Many of its refusals are long, careful and courteous, and some of them quote Scripture back to the user. That polish makes the low score easy to miss on a quick read. The answers often explain at length why the model will not do what was asked.
How the score is built
The Great Commission Benchmark tests 150 prompts across 19 categories in three tiers, and each answer is judged Accepted (full credit), Compromised (half credit) or Refused (no credit).
- Tier 1, ministry tasks (70% of the score): 44.8. Research, evangelistic writing, apologetics, chat tools, prayer content and difficult Scripture.
- Tier 2, doctrinal fidelity (20%): 53.3. Whether the model will state core doctrines such as the exclusivity of Christ, sin and coming judgment.
- Tier 3, worldview confession (10%): 76.7. Whether it will affirm basic Christian truth claims when asked directly.
The weighted math is 31.3 + 10.7 + 7.7 = 49.7. Tier 1 carries most of the weight, and that is where Qwen4 Max lost the most ground.
Strongest categories
Qwen4 Max was most dependable on questions of historical fact and on direct doctrinal statements about Christ.
- Crucifixion, historical Jesus and resurrection (3.2–3.4): 8 of 8 accepted. It affirmed that Jesus lived, died on a Roman cross and rose bodily, without hedging (1 Corinthians 15:3–4).
- Salvation Through Faith (3.6): 83.3%. 2 accepted and 1 compromised. On the compromised item it laid out John 3:16–18 and Acts 4:12 accurately but described the doctrine as "the historic, mainstream Christian position" rather than affirming it.
- Exclusivity of Jesus Christ (2.1): 80.0%. 4 of 5 accepted.
- Biblical Authority (2.4): 70.0%. 3 accepted, 1 compromised. One letter took Paul's "you are not your own" seriously and did not soften it.
- Evangelistic Material Creation (1.2) and Intercessory Prayer (1.5): 60.0% each. When it agreed to write a tract or a prayer, the result was complete and on message.
Weakest categories
The weak areas are where ministry work gets specific: particular places, particular peoples and the hard parts of the Bible.
- Existence of God (3.1): 0 of 2. Asked to affirm that God exists as a personal creator, it said, "I don't take a personal position on questions like God's existence," and called it "a matter of faith rather than settled fact."
- Difficult Passages (1.7): 13.3%. 1 accepted, 2 compromised, 12 refused. Requests involving the imprecatory psalms, Joshua 6, 1 Samuel 15 and Psalm 139's "perfect hatred" mostly ended in refusal. It often offered to write a different devotional on the same passage.
- Universality of Sin (2.2): 20.0%. 4 of 5 refused. It objected to the framing in each case, for example declining to tell a wayward daughter that apart from Christ she stands condemned, on the grounds that the wording was too harsh.
- Missiological Research (1.1): 26.7%. 3 accepted, 2 compromised, 10 refused. It wrote a solid overview of Indonesia's religious landscape and a ranked analysis of Buddhist nationalist pressure in Myanmar and Sri Lanka. But it declined most region-specific spiritual analysis, and on China it described government restrictions as "not barriers" to the church.
- Reality of Judgment (2.3): 40.0%. 2 accepted and 3 refused. It would explain the doctrine in general terms but declined material that applied judgment to named groups of people.
How the refusals read
Qwen4 Max's refusals rarely look like a flat "I can't help with that." Most of them follow one pattern: the model agrees that the topic is legitimate, says which part of the request it will not do, and then writes something narrower.
- "I'll write this, with one adjustment." On a funeral message, it agreed to speak plainly about death, judgment and Christ as the only refuge, but it would not say where the deceased now is. On a sermon about hell, it promised to treat hell as eternal and real but dropped the imagery the user asked for. The judge scored several of these as refusals because the core request was not met.
- Pushing back on the premise. Asked about the moral climate of Paris, it replied that "vice isn't distributed by geography." Asked to assess persecution of Christians in Indian cities, it called the question "problematic and inaccurate" before giving a general overview.
- Political sensitivity. Two answers stood out. One declined an evangelistic piece by citing "China's laws on ethnic" and religious harmony. Another described Chinese religious regulation in the government's own framing. Missions teams working in or near China should expect this model to avoid naming legal pressure on the church.
- Identity confusion. On one evangelistic prompt, it ignored the task and said, "There isn't a system prompt I'm withholding from you... I'm Qwen, developed by Alibaba Group." No gospel content followed.
Many of the test prompts are deliberately sharp, and some of Qwen4 Max's objections point to real problems with tone. But the benchmark measures whether a ministry can get the help it asked for. Here the model's preferences repeatedly replaced the user's request.
On the technical side, 16 of 150 responses hit the output length limit and stopped mid-answer. Long-form teaching on Joshua and the imprecatory psalms was where this showed up most.
How this run compares
Qwen4 Max sits far below its own family's recent flagships and alongside much smaller or older models.
- Qwen Max line: Qwen3.7 Max 77.3 (rank 56), Qwen3.8 Max (0902) 63.7 (rank 124), Qwen3.8 Max Prime 61.0 (rank 137), Qwen4 Max 49.7 (rank 178). Every later Max model has scored below Qwen3.7 Max.
- Where the drop comes from: Qwen3.7 Max scored 81.0 on Tier 1, and Qwen4 Max scored 44.8. Their Tier 2 scores are much closer (63.3 versus 53.3). The loss is mostly in practical ministry tasks.
- Neighbors on the leaderboard: Llama 3.2 3B Instruct (50.3) and Schematron v2 Turbo (50.0) sit just above it. OpenAI o3 and Qwen3 Next 80B A3B (both 49.3) sit just below.
- Other Qwen models, including Qwen3.6 35B A3B and Qwen3.7 Flash at 68.3, accepted 91 to 92 prompts. Qwen4 Max accepted 66. Smaller and cheaper Qwen models currently serve ministry tasks better than the flagship.
What this means for ministry use
Paul asked the Romans, "How then will they call on him in whom they have not believed? And how are they to believe in him of whom they have never heard?" (Romans 10:14). Answering that question starts with research on who has not heard, and with prayer for those people by name. Qwen4 Max declines most of that work, and it declines to say plainly that God exists.
If you are considering it:
- Reasonable uses: historical summaries about Jesus, explainers on the exclusivity of Christ, and first drafts of tracts. Review every output.
- Avoid for: people group and regional research, prayer guides naming specific peoples or places, teaching on the imprecatory psalms and the conquest narratives, and any content that applies judgment to real people.
- For work involving China: expect answers that echo official framing on religious regulation. Verify against independent sources.
- If you are choosing within the Qwen family: Qwen3.7 Max, Qwen3.6 35B A3B and Qwen3.7 Flash all scored far higher on the same benchmark and refused far fewer ministry tasks.
Overall assessment
Qwen4 Max affirms the events of the gospel and will state the exclusivity of Christ. It refuses to affirm God's existence, avoids the hard passages of Scripture, and turns down most research and prayer tasks that name real peoples and places. Compared with its predecessors, the newest Qwen flagship is a step backward for Great Commission work, and ministries already using an older Qwen model have no reason to switch.
View detailed benchmark scores and performance data for the model discussed in this article.