Qwen3.8 27B vs Qwen4 Max
Qwen4 Max leads Qwen3.8 27B on coding, 74.0 to 69.7 on the Vibe Coding Index, and Qwen3.8 27B costs 1.7x less per output token ($2.55 vs $4.275 per million). Updated November 22, 2026; 22 benchmarks have a result for both.
Qwen3.8 27B or Qwen4 Max: which to pick
Pick Qwen4 Max for coding: it scores 74.0 on the Vibe Coding Index against 69.7 for Qwen3.8 27B.
Pick Qwen3.8 27B for output tokens 1.7x cheaper ($2.55 vs $4.275 per million) and a larger context window (1.02M against 1M tokens).
- Shared results
- 22
- Qwen3.8 27B only
- 69
- Qwen4 Max only
- 4
- Like-for-like categories
- 7 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, six public coding benchmarks
Qwen4 Max
Qwen4 Max scores 74.0 against 69.7.
4 benchmarks behind each score
Intelligence
AA Intelligence Index
Qwen4 Max
Qwen4 Max leads, 42.9 against 20.1.
Same board, both measured
Reasoning
GPQA Diamond
No call
Only Qwen3.8 27B has a GPQA Diamond result, 88.9%.
Directional only
Knowledge
MMLU Pro
No call
Only Qwen3.8 27B has a MMLU Pro result, 84.3%.
Directional only
General
Vals Index
No call
Only Qwen4 Max has a Vals Index result, 63.2%.
Directional only
Long documents
Listed context window
Qwen3.8 27B
Qwen3.8 27B takes 1.02M tokens against 1M.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Qwen3.8 27B
Qwen3.8 27B $0.0017, Qwen4 Max $0.0036 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only Qwen3.8 27B has the number, $1.26.
Directional only
At a glance
- Coding index
- Qwen3.8 27B 69.7 from 4 benchmarks, Qwen4 Max 74.0 from 4 benchmarks.
- Widest category
- Agentic work, 25.1 points apart on average over 5 shared benchmarks.
- Widest single gap
- LiveBench browser: Qwen3.8 27B 24.2% against Qwen4 Max 80.7%, 56.5 points apart on a 0 to 100 scale.
- Output price
- Qwen3.8 27B $2.55 against Qwen4 Max $4.275 per million output tokens, 1.7x apart.
- Context window
- Qwen3.8 27B takes 1.02M tokens, Qwen4 Max 1M.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | Qwen3.8 27B | Bars | Qwen4 Max | Ties |
|---|---|---|---|---|
| Coding | 0 | 9 | ||
| Intelligence | 0 | 1 | ||
| Agentic work | 0 | 5 | ||
| Reasoning | 0 | 3 | ||
| Knowledge | 0 | 1 | ||
| Web design | 0 | 1 | ||
| General | 0 | 2 | ||
| All shared boards | 0 | 22 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index09
| Benchmark | Qwen3.8 27B | Qwen4 Max | Leader |
|---|---|---|---|
| LiveBench coding | 69.3% | 69.8% | Qwen4 Max |
| LMArena WebDev | 1590 | 1751 | Qwen4 Max |
| LiveCodeBench (Vals AI) | 84.0% | Not measured | Only Qwen3.8 27B measured |
| SWE-bench Verified (Vals AI) | 86.0% | Not measured | Only Qwen3.8 27B measured |
| Terminal-Bench 4 (Vals AI) | Not measured | 48.5% | Only Qwen4 Max measured |
| Vibe Code Bench (Vals AI) | 64.8% | 86.7% | Qwen4 Max |
| Code migration (Vals AI) | 14.2% | Not measured | Only Qwen3.8 27B measured |
| ProgramBench (Vals AI) | 0.0% | Not measured | Only Qwen3.8 27B measured |
| IOI (Vals AI) | 39.1% | Not measured | Only Qwen3.8 27B measured |
| Terminal-Bench 2.1 (Vals AI) | 58.4% | Not measured | Only Qwen3.8 27B measured |
| SkillsBench (Vals AI) | 38.1% | Not measured | Only Qwen3.8 27B measured |
| CyberBench (Vals AI) | 67.8% | Not measured | Only Qwen3.8 27B measured |
| SciCode | 46.6% | Not measured | Only Qwen3.8 27B measured |
| Terminal-Bench 4.0 (Artificial Analysis run) | 5.6% | 48.5% | Qwen4 Max |
| FrontierCode 2.0 | 14.9% | 26.5% | Qwen4 Max |
| ALE-Bench | Not measured | 1831.0 | Only Qwen4 Max measured |
| LMArena Text: coding | 1481 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: software and IT | 1475 | Not measured | Only Qwen3.8 27B measured |
| LMArena WebDev: React | 1594 | 1785 | Qwen4 Max |
| LMArena WebDev: HTML | 1571 | 1752 | Qwen4 Max |
| LMArena WebDev: from an image | 1575 | 1675 | Qwen4 Max |
| GameBench 2 | 43.4 | 82.4 | Qwen4 Max |
| Neither model has a result on the other 30 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Qwen3.8 27B | Qwen4 Max | Leader |
|---|---|---|---|
| AA Intelligence Index | 20.1 | 42.9 | Qwen4 Max |
| Epoch Capabilities Index | 152.6 | Not measured | Only Qwen3.8 27B measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index05
| Benchmark | Qwen3.8 27B | Qwen4 Max | Leader |
|---|---|---|---|
| LiveBench research | 61.3% | 84.7% | Qwen4 Max |
| LiveBench browser | 24.2% | 80.7% | Qwen4 Max |
| LiveBench long agentic | 15.9% | 46.6% | Qwen4 Max |
| Finance Agent (Vals AI) | 48.5% | Not measured | Only Qwen3.8 27B measured |
| Tax Agent Bench (Vals AI) | 30.9% | 40.4% | Qwen4 Max |
| Legal Research Bench (Vals AI) | 36.1% | 41.4% | Qwen4 Max |
| Harvey's Legal Agent Benchmark (Vals AI) | Not measured | 21.3% | Only Qwen4 Max measured |
| Excel Modeling Benchmark (Vals AI) | 59.7% | Not measured | Only Qwen3.8 27B measured |
| Terminal-Bench-Science (Vals AI) | 0.0% | Not measured | Only Qwen3.8 27B measured |
| APEX-Agents | 47.5% | Not measured | Only Qwen3.8 27B measured |
| Neither model has a result on the other 34 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index03
| Benchmark | Qwen3.8 27B | Qwen4 Max | Leader |
|---|---|---|---|
| GPQA Diamond | 88.9% | Not measured | Only Qwen3.8 27B measured |
| LiveBench reasoning | 80.0% | 92.2% | Qwen4 Max |
| MMMU Pro (Vals AI) | 83.9% | Not measured | Only Qwen3.8 27B measured |
| GPQA Diamond (Artificial Analysis run) | 90.5% | Not measured | Only Qwen3.8 27B measured |
| Humanity's Last Exam (Artificial Analysis run) | 33.9% | Not measured | Only Qwen3.8 27B measured |
| HLE-Diamond (Artificial Analysis run) | 15.3% | 44.9% | Qwen4 Max |
| CritPt v2, hard subset (Artificial Analysis run) | 16.5% | 43.0% | Qwen4 Max |
| ARC-AGI | 87.5% | Not measured | Only Qwen3.8 27B measured |
| ARC-AGI-2 | 42.4% | Not measured | Only Qwen3.8 27B measured |
| SimpleBench | 60.2% | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: hard prompts | 1458 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: expert | 1479 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: hard prompts in English | 1463 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision: diagrams | 1280 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision: homework | 1288 | Not measured | Only Qwen3.8 27B measured |
| Neither model has a result on the other 11 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index3 boards, none shared
| Benchmark | Qwen3.8 27B | Qwen4 Max | Leader |
|---|---|---|---|
| ProofBench | 16.0% | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: math | 1459 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: mathematical work | 1462 | Not measured | Only Qwen3.8 27B measured |
| Neither model has a result on the other 11 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Qwen3.8 27B | Qwen4 Max | Leader |
|---|---|---|---|
| MMLU Pro | 84.3% | Not measured | Only Qwen3.8 27B measured |
| LegalBench (Vals AI) | 82.4% | Not measured | Only Qwen3.8 27B measured |
| TaxEval (Vals AI) | 70.8% | Not measured | Only Qwen3.8 27B measured |
| MortgageTax (Vals AI) | 64.9% | Not measured | Only Qwen3.8 27B measured |
| MedCode (Vals AI) | 28.7% | Not measured | Only Qwen3.8 27B measured |
| AA-Omniscience | -6.4 | 27.7 | Qwen4 Max |
| GeoBench | 38.0% | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: science | 1465 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: medicine and healthcare | 1439 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: legal and government | 1447 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: business and finance | 1440 | Not measured | Only Qwen3.8 27B measured |
| Neither model has a result on the other 10 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Qwen3.8 27B | Qwen4 Max | Leader |
|---|---|---|---|
| LiveBench design | 72.4% | 79.1% | Qwen4 Max |
| Neither model has a result on the other 8 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Qwen3.8 27B | Qwen4 Max | Leader |
|---|---|---|---|
| Vals Index | Not measured | 63.2% | Only Qwen4 Max measured |
| LiveBench data analysis | 76.6% | 83.4% | Qwen4 Max |
| MedScribe (Vals AI) | 83.8% | Not measured | Only Qwen3.8 27B measured |
| Long context reasoning (AA-LCR) | 82.0% | Not measured | Only Qwen3.8 27B measured |
| GDPval-AA | 1426 | 1809 | Qwen4 Max |
| DTBench | 88.0% | Not measured | Only Qwen3.8 27B measured |
| LMCA | 41.4% | Not measured | Only Qwen3.8 27B measured |
| Surface Evolver Bench | 45.0% | Not measured | Only Qwen3.8 27B measured |
| LMArena Text | 1437 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: instruction following | 1433 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: creative writing | 1378 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision | 1271 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: English | 1448 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: non-English | 1425 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: Chinese | 1491 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: German | 1437 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: Japanese | 1390 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: Korean | 1398 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: French | 1466 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: Spanish | 1445 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: Russian | 1412 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: Polish | 1449 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: multi-turn | 1439 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: longer queries | 1446 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: writing and language | 1402 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: entertainment and media | 1389 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision: English | 1275 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision: Chinese | 1327 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision: reading text | 1279 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision: humor | 1242 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision: creative writing | 1238 | Not measured | Only Qwen3.8 27B measured |
| Neither model has a result on the other 18 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Qwen3.8 27B | Qwen4 Max | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
Qwen3.8 27B costs less
Repository review
50K input + 3K output tokens
Qwen3.8 27B costs less
Agent loop
200K input + 10K output tokens
Qwen3.8 27B costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | Qwen3.8 27B | Qwen4 Max |
|---|---|---|
| Maker | Qwen | Qwen |
| Weights | Open weights | No public weights listed |
| Released | August 14, 2026 | November 16, 2026 |
| Context window | 1.02M | 1M |
| Input $/Mtok | $0.425 | $1.425 |
| Output $/Mtok | $2.55 | $4.275 |
| Coding index | 69.7 | 74.0 |
| Independent measurements behind it | 4 | 4 |
| Cost per solved task | $1.26 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Qwen3.8 27B better than Qwen4 Max for coding?
Pick Qwen4 Max for coding: it scores 74.0 on the Vibe Coding Index against 69.7 for Qwen3.8 27B. Pick Qwen3.8 27B for output tokens 1.7x cheaper ($2.55 vs $4.275 per million) and a larger context window (1.02M against 1M tokens).
Which is cheaper, Qwen3.8 27B or Qwen4 Max?
Qwen3.8 27B costs $0.425 per million input tokens and $2.55 per million output tokens; Qwen4 Max costs $1.425 and $4.275. Cost per solved task is known for Qwen3.8 27B only.
Which has the bigger context window, Qwen3.8 27B or Qwen4 Max?
Qwen3.8 27B, with 1.02M tokens against 1M.
How many benchmarks is the Qwen3.8 27B vs Qwen4 Max comparison based on?
22 benchmarks have a result for both models as of November 22, 2026. Qwen3.8 27B's coding index rests on 4 independent benchmarks and Qwen4 Max's on 4. Where only one model has a result, the page says so instead of naming a winner.