GPT-6 Luna vs Qwen3 235B A22B Thinking 2507
GPT-6 Luna leads Qwen3 235B A22B Thinking 2507 on the Artificial Analysis Intelligence Index, 38.1 to 12.7, and GPT-6 Luna costs 7.0x less per output token ($0.50 vs $3.50 per million). Updated October 22, 2026; 40 benchmarks have a result for both.
GPT-6 Luna or Qwen3 235B A22B Thinking 2507: which to pick
Only GPT-6 Luna has a coding score, 57.3 on the Vibe Coding Index; Qwen3 235B A22B Thinking 2507 has none, so the two cannot be ranked on coding.
Pick Qwen3 235B A22B Thinking 2507 for work measured by LMArena Text: software and IT or Mystery game puzzles, where it leads.
- Shared results
- 40
- GPT-6 Luna only
- 64
- Qwen3 235B A22B Thinking 2507 only
- 6
- 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
No call
Only GPT-6 Luna has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
GPT-6 Luna
GPT-6 Luna leads, 38.1 against 12.7.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only Qwen3 235B A22B Thinking 2507 has a AA Agentic Index result, 1.3.
Directional only
Web design
Design Arena: websites
GPT-6 Luna
GPT-6 Luna leads, 1271 against 1052.
Same board, both measured
General
Vals Index
No call
Only GPT-6 Luna has a Vals Index result, 52.3%.
Directional only
Long documents
Listed context window
GPT-6 Luna
GPT-6 Luna takes 1.05M tokens against 128K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
GPT-6 Luna
GPT-6 Luna $0.0003, Qwen3 235B A22B Thinking 2507 $0.0022 per turn.
List prices
At a glance
- Coding index
- GPT-6 Luna 57.3 from 5 benchmarks, Qwen3 235B A22B Thinking 2507 not benchmarked.
- Widest category
- Web design, 33.3 points apart on average over 6 shared benchmarks.
- Widest single gap
- Design Arena: UI components: GPT-6 Luna 1295 against Qwen3 235B A22B Thinking 2507 942, 44.1 points apart on a 0 to 100 scale.
- Output price
- GPT-6 Luna $0.50 against Qwen3 235B A22B Thinking 2507 $3.50 per million output tokens, 7.0x apart.
- Context window
- GPT-6 Luna takes 1.05M tokens, Qwen3 235B A22B Thinking 2507 128K.
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-6 Luna | Bars | Qwen3 235B A22B Thinking 2507 | Ties |
|---|---|---|---|---|
| Coding | 1 | 1 | ||
| Intelligence | 1 | 0 | ||
| Reasoning | 1 | 4 | ||
| Math | 1 | 2 | ||
| Knowledge | 1 | 4 | ||
| Web design | 6 | 0 | ||
| General | 5 | 13 | ||
| All shared boards | 16 | 24 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index11
| Benchmark | GPT-6 Luna | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| LiveBench coding | 84.5% | Not measured | Only GPT-6 Luna measured |
| LMArena WebDev | 1578 | Not measured | Only GPT-6 Luna measured |
| AA Coding Index | Not measured | 22.1 | Only Qwen3 235B A22B Thinking 2507 measured |
| Terminal-Bench 4 (Vals AI) | 13.6% | Not measured | Only GPT-6 Luna measured |
| Vibe Code Bench (Vals AI) | 81.7% | Not measured | Only GPT-6 Luna measured |
| Code migration (Vals AI) | 42.5% | Not measured | Only GPT-6 Luna measured |
| ProgramBench (Vals AI) | 0.5% | Not measured | Only GPT-6 Luna measured |
| IOI (Vals AI) | 55.6% | Not measured | Only GPT-6 Luna measured |
| Terminal-Bench 2.1 (Vals AI) | 73.0% | Not measured | Only GPT-6 Luna measured |
| SWE-bench Multimodal (Vals AI) | 37.5% | Not measured | Only GPT-6 Luna measured |
| CyberBench (Vals AI) | 76.3% | Not measured | Only GPT-6 Luna measured |
| SciCode | 54.6% | Not measured | Only GPT-6 Luna measured |
| FrontierCode | 42.4% | Not measured | Only GPT-6 Luna measured |
| WeirdML | Not measured | 41.0% | Only Qwen3 235B A22B Thinking 2507 measured |
| ALE-Bench | 1576.9 | Not measured | Only GPT-6 Luna measured |
| LMArena Text: coding | 1437 | 1424 | GPT-6 Luna |
| LMArena Text: software and IT | 1423 | 1431 | Qwen3 235B A22B Thinking 2507 |
| LMArena WebDev: React | 1573 | Not measured | Only GPT-6 Luna measured |
| LMArena WebDev: HTML | 1595 | Not measured | Only GPT-6 Luna measured |
| LMArena WebDev: from an image | 1571 | Not measured | Only GPT-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 index10
| Benchmark | GPT-6 Luna | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| AA Intelligence Index | 38.1 | 12.7 | GPT-6 Luna |
| Epoch Capabilities Index | 156.8 | Not measured | Only GPT-6 Luna measured |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index19 boards, none shared
| Benchmark | GPT-6 Luna | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| AA Agentic Index | Not measured | 1.3 | Only Qwen3 235B A22B Thinking 2507 measured |
| Design Arena: agents building web apps | 1191 | Not measured | Only GPT-6 Luna measured |
| Design Arena: agents building full-stack apps | 1158 | Not measured | Only GPT-6 Luna measured |
| Design Arena: agents building mobile apps | 1131 | Not measured | Only GPT-6 Luna measured |
| LiveBench research | 43.5% | Not measured | Only GPT-6 Luna measured |
| LiveBench browser | 72.0% | Not measured | Only GPT-6 Luna measured |
| LiveBench long agentic | 48.1% | Not measured | Only GPT-6 Luna measured |
| SREBench (Vals AI) | 2.7% | Not measured | Only GPT-6 Luna measured |
| Finance Agent (Vals AI) | 49.9% | Not measured | Only GPT-6 Luna measured |
| Tax Agent Bench (Vals AI) | 19.9% | Not measured | Only GPT-6 Luna measured |
| Legal Research Bench (Vals AI) | 30.3% | Not measured | Only GPT-6 Luna measured |
| Harvey's Legal Agent Benchmark (Vals AI) | 4.2% | Not measured | Only GPT-6 Luna measured |
| Excel Modeling Benchmark (Vals AI) | 68.5% | Not measured | Only GPT-6 Luna measured |
| BioMysteryBench (Vals AI) | 61.5% | Not measured | Only GPT-6 Luna measured |
| Terminal-Bench-Science (Vals AI) | 4.3% | Not measured | Only GPT-6 Luna measured |
| Vals Science Index | 28.4% | Not measured | Only GPT-6 Luna measured |
| APEX-Agents | 44.3% | Not measured | Only GPT-6 Luna measured |
| SpireClimb | 0.0% | Not measured | Only GPT-6 Luna measured |
| Vending-Bench 2 | Not measured | -$11 | Only Qwen3 235B A22B Thinking 2507 measured |
| Neither model has a result on the other 23 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index14
| Benchmark | GPT-6 Luna | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| LiveBench reasoning | 81.8% | Not measured | Only GPT-6 Luna measured |
| MysteryMechanism (Vals AI) | 19.4% | Not measured | Only GPT-6 Luna measured |
| Humanity's Last Exam (Artificial Analysis run) | 38.5% | Not measured | Only GPT-6 Luna measured |
| GPQA Diamond (Epoch AI run) | 90.5% | 80.1% | GPT-6 Luna |
| ARC-AGI | 86.7% | Not measured | Only GPT-6 Luna measured |
| ARC-AGI-2 | 59.3% | Not measured | Only GPT-6 Luna measured |
| Chess puzzles | 36.0% | Not measured | Only GPT-6 Luna measured |
| Mystery game puzzles | 7.0% | 9.0% | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: hard prompts | 1410 | 1416 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: expert | 1442 | 1463 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: hard prompts in English | 1410 | 1432 | Qwen3 235B A22B Thinking 2507 |
| LMArena Vision: diagrams | 1245 | Not measured | Only GPT-6 Luna measured |
| LMArena Vision: homework | 1261 | Not measured | Only GPT-6 Luna measured |
| Neither model has a result on the other 10 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index12
| Benchmark | GPT-6 Luna | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| OTIS Mock AIME | 98.9% | 86.7% | GPT-6 Luna |
| FrontierMath, tiers 1 to 3 | 78.9% | Not measured | Only GPT-6 Luna measured |
| FrontierMath, tier 4 | 56.1% | Not measured | Only GPT-6 Luna measured |
| ProofBench | 64.0% | Not measured | Only GPT-6 Luna measured |
| LMArena Text: math | 1409 | 1412 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: mathematical work | 1427 | 1435 | Qwen3 235B A22B Thinking 2507 |
| Neither model has a result on the other 8 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index14
| Benchmark | GPT-6 Luna | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| MedCode (Vals AI) | 44.7% | Not measured | Only GPT-6 Luna measured |
| SimpleQA Verified | 41.4% | 40.4% | GPT-6 Luna |
| LMArena Text: science | 1386 | 1433 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: medicine and healthcare | 1352 | 1456 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: legal and government | 1400 | 1415 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: business and finance | 1388 | 1413 | Qwen3 235B A22B Thinking 2507 |
| FACTS Benchmark Suite | 63.3% | Not measured | Only GPT-6 Luna measured |
| PRBench Legal | 42.9% | Not measured | Only GPT-6 Luna measured |
| Neither model has a result on the other 12 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index60
| Benchmark | GPT-6 Luna | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| Design Arena: websites | 1271 | 1052 | GPT-6 Luna |
| Design Arena: UI components | 1295 | 942 | GPT-6 Luna |
| Design Arena: game dev | 1258 | 956 | GPT-6 Luna |
| Design Arena: data viz | 1217 | 950 | GPT-6 Luna |
| Design Arena: 3D | 1218 | 997 | GPT-6 Luna |
| Design Arena: all code categories | 1270 | 1032 | GPT-6 Luna |
| Design Arena: SVG | 1253 | Not measured | Only GPT-6 Luna measured |
| Design Arena: ASCII art | 1153 | Not measured | Only GPT-6 Luna measured |
| LiveBench design | 82.5% | Not measured | Only GPT-6 Luna measured |
GeneralLike-for-likePublic boards, shown as published, kept out of the index513
| Benchmark | GPT-6 Luna | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| Vals Index | 52.3% | Not measured | Only GPT-6 Luna measured |
| LiveBench data analysis | 73.4% | Not measured | Only GPT-6 Luna measured |
| Public Benefits Bench (Vals AI) | 57.6% | Not measured | Only GPT-6 Luna measured |
| SAGE (Vals AI) | 48.1% | Not measured | Only GPT-6 Luna measured |
| MedScribe (Vals AI) | 83.7% | Not measured | Only GPT-6 Luna measured |
| Long context reasoning (AA-LCR) | 83.3% | Not measured | Only GPT-6 Luna measured |
| DTBench | 90.1% | 80.3% | GPT-6 Luna |
| LMCA | 44.5% | 29.3% | GPT-6 Luna |
| Fiction.LiveBench | Not measured | 75.0% | Only Qwen3 235B A22B Thinking 2507 measured |
| Creative writing (Lech Mazur) | Not measured | 82.4% | Only Qwen3 235B A22B Thinking 2507 measured |
| Furniture assembly | 44.2% | Not measured | Only GPT-6 Luna measured |
| LMArena Text | 1389 | 1415 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: instruction following | 1408 | 1386 | GPT-6 Luna |
| LMArena Text: creative writing | 1363 | 1384 | Qwen3 235B A22B Thinking 2507 |
| LMArena Vision | 1216 | Not measured | Only GPT-6 Luna measured |
| LMArena Text: English | 1393 | 1429 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: non-English | 1381 | 1397 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: Chinese | 1427 | 1481 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: German | 1376 | 1408 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: Japanese | 1369 | 1386 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: Korean | 1359 | 1375 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: French | 1408 | Not measured | Only GPT-6 Luna measured |
| LMArena Text: Spanish | 1391 | 1389 | GPT-6 Luna |
| LMArena Text: Russian | 1392 | 1401 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: Polish | 1383 | 1390 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: multi-turn | 1392 | 1406 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: longer queries | 1408 | 1400 | GPT-6 Luna |
| LMArena Text: writing and language | 1384 | 1388 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: entertainment and media | 1364 | 1370 | Qwen3 235B A22B Thinking 2507 |
| LMArena Vision: English | 1204 | Not measured | Only GPT-6 Luna measured |
| LMArena Vision: Chinese | 1266 | Not measured | Only GPT-6 Luna measured |
| LMArena Vision: reading text | 1225 | Not measured | Only GPT-6 Luna measured |
| LMArena Vision: creative writing | 1205 | Not measured | Only GPT-6 Luna measured |
| Neither model has a result on the other 15 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-6 Luna | Qwen3 235B A22B Thinking 2507 | 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-6 Luna costs less
Repository review
50K input + 3K output tokens
GPT-6 Luna costs less
Agent loop
200K input + 10K output tokens
GPT-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-6 Luna | Qwen3 235B A22B Thinking 2507 |
|---|---|---|
| Maker | OpenAI | Qwen |
| Weights | No public weights listed | Open weights |
| Released | September 22, 2026 | July 25, 2025 |
| Context window | 1.05M | 128K |
| Input $/Mtok | $0.10 | $0.45 |
| Output $/Mtok | $0.50 | $3.50 |
| Coding index | 57.3 | Not benchmarked |
| Independent measurements behind it | 5 | 0 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GPT-6 Luna better than Qwen3 235B A22B Thinking 2507 for coding?
Only GPT-6 Luna has a coding score, 57.3 on the Vibe Coding Index; Qwen3 235B A22B Thinking 2507 has none, so the two cannot be ranked on coding. Pick Qwen3 235B A22B Thinking 2507 for work measured by LMArena Text: software and IT or Mystery game puzzles, where it leads.
Which is cheaper, GPT-6 Luna or Qwen3 235B A22B Thinking 2507?
GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens; Qwen3 235B A22B Thinking 2507 costs $0.45 and $3.50. Cost per solved task is known for neither of them.
Which has the bigger context window, GPT-6 Luna or Qwen3 235B A22B Thinking 2507?
GPT-6 Luna, with 1.05M tokens against 128K.
How many benchmarks is the GPT-6 Luna vs Qwen3 235B A22B Thinking 2507 comparison based on?
40 benchmarks have a result for both models as of October 22, 2026. GPT-6 Luna's coding index rests on 5 independent benchmarks and Qwen3 235B A22B Thinking 2507's on 0. Where only one model has a result, the page says so instead of naming a winner.