GPT-5.6 Luna vs Qwen4 Max
Qwen4 Max leads GPT-5.6 Luna on coding, 74.0 to 55.3 on the Vibe Coding Index, and GPT-5.6 Luna costs 3.6x less per output token ($1.20 vs $4.275 per million). Updated November 22, 2026; 17 benchmarks have a result for both.
GPT-5.6 Luna or Qwen4 Max: which to pick
Pick Qwen4 Max for coding: it scores 74.0 on the Vibe Coding Index against 55.3 for GPT-5.6 Luna, though on 4 benchmarks against 5.
Pick GPT-5.6 Luna for output tokens 3.6x cheaper ($1.20 vs $4.275 per million) and a larger context window (1.05M against 1M tokens).
- Shared results
- 17
- GPT-5.6 Luna only
- 105
- Qwen4 Max only
- 9
- Like-for-like categories
- 6 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 55.3.
5 and 4 benchmarks behind the scores
Intelligence
AA Intelligence Index
Qwen4 Max
Qwen4 Max leads, 42.9 against 23.8.
Same board, both measured
Reasoning
GPQA Diamond
No call
Only GPT-5.6 Luna has a GPQA Diamond result, 91.7%.
Directional only
Knowledge
MMLU Pro
No call
Only GPT-5.6 Luna has a MMLU Pro result, 86.0%.
Directional only
Web design
Design Arena: websites
No call
Only GPT-5.6 Luna has a Design Arena: websites result, 1244.
Directional only
General
Vals Index
No call
Only Qwen4 Max has a Vals Index result, 63.2%.
Directional only
Long documents
Listed context window
GPT-5.6 Luna
GPT-5.6 Luna takes 1.05M tokens against 1M.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
GPT-5.6 Luna
GPT-5.6 Luna $0.0008, Qwen4 Max $0.0036 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only GPT-5.6 Luna has the number, $0.04.
Directional only
At a glance
- Coding index
- GPT-5.6 Luna 55.3 from 5 benchmarks, Qwen4 Max 74.0 from 4 benchmarks.
- Widest category
- General, 28.2 points apart on average over 1 shared benchmark.
- Widest single gap
- Terminal-Bench 4 (Vals AI): GPT-5.6 Luna 11.6% against Qwen4 Max 48.5%, 49.1 points apart on a 0 to 100 scale.
- Output price
- GPT-5.6 Luna $1.20 against Qwen4 Max $4.275 per million output tokens, 3.6x apart.
- Context window
- GPT-5.6 Luna takes 1.05M 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 | GPT-5.6 Luna | Bars | Qwen4 Max | Ties |
|---|---|---|---|---|
| Coding | 0 | 10 | ||
| Intelligence | 0 | 1 | ||
| Agentic work | 0 | 2 | ||
| Reasoning | 0 | 2 | ||
| Knowledge | 0 | 1 | ||
| General | 0 | 1 | ||
| All shared boards | 0 | 17 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index010
| Benchmark | GPT-5.6 Luna | Qwen4 Max | Leader |
|---|---|---|---|
| LiveBench coding | Not measured | 69.8% | Only Qwen4 Max measured |
| LMArena WebDev | 1519 | 1751 | Qwen4 Max |
| SWE-bench Verified (Vals AI) | 93.0% | Not measured | Only GPT-5.6 Luna measured |
| Terminal-Bench 4 (Vals AI) | 11.6% | 48.5% | Qwen4 Max |
| Vibe Code Bench (Vals AI) | 77.1% | 86.7% | Qwen4 Max |
| Code migration (Vals AI) | 44.5% | Not measured | Only GPT-5.6 Luna measured |
| ProgramBench (Vals AI) | 0.0% | Not measured | Only GPT-5.6 Luna measured |
| IOI (Vals AI) | 61.8% | Not measured | Only GPT-5.6 Luna measured |
| Terminal-Bench 2.1 (Vals AI) | 79.0% | Not measured | Only GPT-5.6 Luna measured |
| Vibe Code Bench 1-100 (Vals AI) | 22.6% | Not measured | Only GPT-5.6 Luna measured |
| SkillsBench (Vals AI) | 60.5% | Not measured | Only GPT-5.6 Luna measured |
| CyberBench (Vals AI) | 73.6% | Not measured | Only GPT-5.6 Luna measured |
| SciCode | 53.6% | Not measured | Only GPT-5.6 Luna measured |
| Terminal-Bench 4.0 (Artificial Analysis run) | 11.6% | 48.5% | Qwen4 Max |
| FrontierCode 2.0 | 17.6% | 26.5% | Qwen4 Max |
| FrontierCode | 39.8% | Not measured | Only GPT-5.6 Luna measured |
| WeirdML | 60.9% | Not measured | Only GPT-5.6 Luna measured |
| MirrorCode | 21.1% | Not measured | Only GPT-5.6 Luna measured |
| ALE-Bench | 1667.4 | 1831.0 | Qwen4 Max |
| LMArena Text: coding | 1467 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: software and IT | 1464 | Not measured | Only GPT-5.6 Luna measured |
| LMArena WebDev: React | 1508 | 1785 | Qwen4 Max |
| LMArena WebDev: HTML | 1522 | 1752 | Qwen4 Max |
| LMArena WebDev: from an image | 1517 | 1675 | Qwen4 Max |
| SciCode | 6.2% | Not measured | Only GPT-5.6 Luna measured |
| GameBench 2 | 53.7 | 82.4 | Qwen4 Max |
| Terminal-Bench 2.1 | 75.7% | Not measured | Only GPT-5.6 Luna measured |
| Terminal-Bench 3.0 | 14.3% | Not measured | Only GPT-5.6 Luna measured |
| Terminal-Bench 4.0 | 17.3% | Not measured | Only GPT-5.6 Luna measured |
| Neither model has a result on the other 23 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | GPT-5.6 Luna | Qwen4 Max | Leader |
|---|---|---|---|
| AA Intelligence Index | 23.8 | 42.9 | Qwen4 Max |
| Epoch Capabilities Index | 156.7 | Not measured | Only GPT-5.6 Luna measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | GPT-5.6 Luna | Qwen4 Max | Leader |
|---|---|---|---|
| LiveBench research | Not measured | 84.7% | Only Qwen4 Max measured |
| LiveBench browser | Not measured | 80.7% | Only Qwen4 Max measured |
| LiveBench long agentic | Not measured | 46.6% | Only Qwen4 Max measured |
| SREBench (Vals AI) | 1.5% | Not measured | Only GPT-5.6 Luna measured |
| Finance Agent (Vals AI) | 55.0% | Not measured | Only GPT-5.6 Luna measured |
| Tax Agent Bench (Vals AI) | 24.8% | 40.4% | Qwen4 Max |
| Legal Research Bench (Vals AI) | 36.5% | 41.4% | Qwen4 Max |
| Harvey's Legal Agent Benchmark (Vals AI) | Not measured | 21.3% | Only Qwen4 Max measured |
| Excel Modeling Benchmark (Vals AI) | 67.1% | Not measured | Only GPT-5.6 Luna measured |
| BioMysteryBench (Vals AI) | 61.5% | Not measured | Only GPT-5.6 Luna measured |
| Terminal-Bench-Science (Vals AI) | 0.0% | Not measured | Only GPT-5.6 Luna measured |
| Vals Science Index | 25.3% | Not measured | Only GPT-5.6 Luna measured |
| APEX-Agents | 43.0% | Not measured | Only GPT-5.6 Luna measured |
| BALROG | 45.6% | Not measured | Only GPT-5.6 Luna measured |
| Vending-Bench 2 | $4,095 | Not measured | Only GPT-5.6 Luna measured |
| Neither model has a result on the other 29 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | GPT-5.6 Luna | Qwen4 Max | Leader |
|---|---|---|---|
| GPQA Diamond | 91.7% | Not measured | Only GPT-5.6 Luna measured |
| LiveBench reasoning | Not measured | 92.2% | Only Qwen4 Max measured |
| MysteryMechanism (Vals AI) | 14.4% | Not measured | Only GPT-5.6 Luna measured |
| MMMU Pro (Vals AI) | 85.0% | Not measured | Only GPT-5.6 Luna measured |
| GPQA Diamond (Artificial Analysis run) | 91.1% | Not measured | Only GPT-5.6 Luna measured |
| Humanity's Last Exam (Artificial Analysis run) | 39.5% | Not measured | Only GPT-5.6 Luna measured |
| HLE-Diamond (Artificial Analysis run) | 20.9% | 44.9% | Qwen4 Max |
| CritPt v2, hard subset (Artificial Analysis run) | 30.5% | 43.0% | Qwen4 Max |
| GPQA Diamond (Epoch AI run) | 91.6% | Not measured | Only GPT-5.6 Luna measured |
| ARC-AGI | 88.0% | Not measured | Only GPT-5.6 Luna measured |
| ARC-AGI-2 | 59.5% | Not measured | Only GPT-5.6 Luna measured |
| SimpleBench | 46.8% | Not measured | Only GPT-5.6 Luna measured |
| Chess puzzles | 43.0% | Not measured | Only GPT-5.6 Luna measured |
| Mystery game puzzles | 21.0% | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: hard prompts | 1452 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: expert | 1477 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: hard prompts in English | 1459 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Vision: diagrams | 1272 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Vision: homework | 1283 | Not measured | Only GPT-5.6 Luna measured |
| Kagi LLM Benchmark | 49.1% | Not measured | Only GPT-5.6 Luna measured |
| GoBench | 1090 | Not measured | Only GPT-5.6 Luna measured |
| Neither model has a result on the other 5 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index6 boards, none shared
| Benchmark | GPT-5.6 Luna | Qwen4 Max | Leader |
|---|---|---|---|
| OTIS Mock AIME | 98.3% | Not measured | Only GPT-5.6 Luna measured |
| FrontierMath, tiers 1 to 3 | 82.1% | Not measured | Only GPT-5.6 Luna measured |
| FrontierMath, tier 4 | 61.0% | Not measured | Only GPT-5.6 Luna measured |
| ProofBench | 60.0% | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: math | 1457 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: mathematical work | 1468 | Not measured | Only GPT-5.6 Luna measured |
| Neither model has a result on the other 8 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | GPT-5.6 Luna | Qwen4 Max | Leader |
|---|---|---|---|
| MMLU Pro | 86.0% | Not measured | Only GPT-5.6 Luna measured |
| LegalBench (Vals AI) | 84.0% | Not measured | Only GPT-5.6 Luna measured |
| CorpFin (Vals AI) | 64.2% | Not measured | Only GPT-5.6 Luna measured |
| TaxEval (Vals AI) | 76.2% | Not measured | Only GPT-5.6 Luna measured |
| MortgageTax (Vals AI) | 67.3% | Not measured | Only GPT-5.6 Luna measured |
| MedCode (Vals AI) | 42.4% | Not measured | Only GPT-5.6 Luna measured |
| AA-Omniscience | -17.1 | 27.7 | Qwen4 Max |
| SimpleQA Verified | 41.0% | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: science | 1434 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: medicine and healthcare | 1399 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: legal and government | 1444 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: business and finance | 1439 | Not measured | Only GPT-5.6 Luna measured |
| FACTS Benchmark Suite | 56.9% | Not measured | Only GPT-5.6 Luna measured |
| Neither model has a result on the other 8 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index9 boards, none shared
| Benchmark | GPT-5.6 Luna | Qwen4 Max | Leader |
|---|---|---|---|
| Design Arena: websites | 1244 | Not measured | Only GPT-5.6 Luna measured |
| Design Arena: UI components | 1232 | Not measured | Only GPT-5.6 Luna measured |
| Design Arena: game dev | 1172 | Not measured | Only GPT-5.6 Luna measured |
| Design Arena: data viz | 1200 | Not measured | Only GPT-5.6 Luna measured |
| Design Arena: 3D | 1086 | Not measured | Only GPT-5.6 Luna measured |
| Design Arena: all code categories | 1212 | Not measured | Only GPT-5.6 Luna measured |
| Design Arena: SVG | 1176 | Not measured | Only GPT-5.6 Luna measured |
| Design Arena: ASCII art | 1135 | Not measured | Only GPT-5.6 Luna measured |
| LiveBench design | Not measured | 79.1% | Only Qwen4 Max measured |
GeneralLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | GPT-5.6 Luna | Qwen4 Max | Leader |
|---|---|---|---|
| Vals Index | Not measured | 63.2% | Only Qwen4 Max measured |
| LiveBench data analysis | Not measured | 83.4% | Only Qwen4 Max measured |
| Public Benefits Bench (Vals AI) | 61.2% | Not measured | Only GPT-5.6 Luna measured |
| MedScribe (Vals AI) | 84.4% | Not measured | Only GPT-5.6 Luna measured |
| Vals Multimodal Index | 69.1% | Not measured | Only GPT-5.6 Luna measured |
| Long context reasoning (AA-LCR) | 83.7% | Not measured | Only GPT-5.6 Luna measured |
| GDPval-AA | 1470 | 1809 | Qwen4 Max |
| DTBench | 89.1% | Not measured | Only GPT-5.6 Luna measured |
| LMCA | 48.5% | Not measured | Only GPT-5.6 Luna measured |
| Surface Evolver Bench | 61.9% | Not measured | Only GPT-5.6 Luna measured |
| Furniture assembly | 42.5% | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text | 1431 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: instruction following | 1437 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: creative writing | 1396 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Vision | 1256 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: English | 1441 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: non-English | 1417 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: Chinese | 1472 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: German | 1456 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: Japanese | 1418 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: Korean | 1409 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: French | 1454 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: Spanish | 1448 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: Russian | 1427 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: Polish | 1404 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: multi-turn | 1435 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: longer queries | 1436 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: writing and language | 1418 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Text: entertainment and media | 1401 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Vision: English | 1260 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Vision: Chinese | 1301 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Vision: reading text | 1263 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Vision: humor | 1193 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Vision: creative writing | 1256 | Not measured | Only GPT-5.6 Luna measured |
| LMArena Document | 1454 | Not measured | Only GPT-5.6 Luna measured |
| EQ-Bench 4 | 1146 | Not measured | Only GPT-5.6 Luna measured |
| Neither model has a result on the other 13 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-5.6 Luna | 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
GPT-5.6 Luna costs less
Repository review
50K input + 3K output tokens
GPT-5.6 Luna costs less
Agent loop
200K input + 10K output tokens
GPT-5.6 Luna 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 | OpenAI GPT-5.6 Luna | Qwen4 Max |
|---|---|---|
| Maker | OpenAI | Qwen |
| Weights | No public weights listed | No public weights listed |
| Released | July 9, 2026 | November 16, 2026 |
| Context window | 1.05M | 1M |
| Input $/Mtok | $0.20 | $1.425 |
| Output $/Mtok | $1.20 | $4.275 |
| Coding index | 55.3 | 74.0 |
| Independent measurements behind it | 5 | 4 |
| Cost per solved task | $0.04 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GPT-5.6 Luna better than Qwen4 Max for coding?
Pick Qwen4 Max for coding: it scores 74.0 on the Vibe Coding Index against 55.3 for GPT-5.6 Luna, though on 4 benchmarks against 5. Pick GPT-5.6 Luna for output tokens 3.6x cheaper ($1.20 vs $4.275 per million) and a larger context window (1.05M against 1M tokens).
Which is cheaper, GPT-5.6 Luna or Qwen4 Max?
GPT-5.6 Luna costs $0.20 per million input tokens and $1.20 per million output tokens; Qwen4 Max costs $1.425 and $4.275. Cost per solved task is known for GPT-5.6 Luna only.
Which has the bigger context window, GPT-5.6 Luna or Qwen4 Max?
GPT-5.6 Luna, with 1.05M tokens against 1M.
How many benchmarks is the GPT-5.6 Luna vs Qwen4 Max comparison based on?
17 benchmarks have a result for both models as of November 22, 2026. GPT-5.6 Luna's coding index rests on 5 independent benchmarks and Qwen4 Max's on 4. Where only one model has a result, the page says so instead of naming a winner.