Llama 4 Maverick vs Qwen3 235B A22B Thinking 2507
Qwen3 235B A22B Thinking 2507 leads Llama 4 Maverick on the Artificial Analysis Intelligence Index, 12.7 to 9.3, and Llama 4 Maverick costs 3.5x less per output token ($1.00 vs $3.50 per million). Updated October 22, 2026; 41 benchmarks have a result for both.
Llama 4 Maverick or Qwen3 235B A22B Thinking 2507: which to pick
Neither model has a coding score yet, so the two cannot be ranked on coding.
Pick Llama 4 Maverick for output tokens 3.5x cheaper ($1.00 vs $3.50 per million) and a larger context window (524K against 128K tokens).
- Shared results
- 41
- Llama 4 Maverick only
- 44
- Qwen3 235B A22B Thinking 2507 only
- 5
- Like-for-like categories
- 8 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
Neither has a coding score yet.
One side not measured
Intelligence
AA Intelligence Index
Qwen3 235B A22B Thinking 2507
Qwen3 235B A22B Thinking 2507 leads, 12.7 against 9.3.
Same board, both measured
Agentic work
AA Agentic Index
Qwen3 235B A22B Thinking 2507
Qwen3 235B A22B Thinking 2507 leads, 1.3 against 0.6.
Same board, both measured
Reasoning
GPQA Diamond
No call
Only Llama 4 Maverick has a GPQA Diamond result, 69.4%.
Directional only
Math
AIME
No call
Only Llama 4 Maverick has a AIME result, 25.2%.
Directional only
Knowledge
MMLU Pro
No call
Only Llama 4 Maverick has a MMLU Pro result, 79.4%.
Directional only
Web design
Design Arena: websites
Qwen3 235B A22B Thinking 2507
Qwen3 235B A22B Thinking 2507 leads, 1052 against 871.
Same board, both measured
Long documents
Listed context window
Llama 4 Maverick
Llama 4 Maverick takes 524K tokens against 128K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Llama 4 Maverick
Llama 4 Maverick $0.0008, Qwen3 235B A22B Thinking 2507 $0.0022 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only Llama 4 Maverick has the number, $1.491.
Directional only
At a glance
- Coding index
- Llama 4 Maverick not benchmarked, Qwen3 235B A22B Thinking 2507 not benchmarked.
- Widest category
- Reasoning, 26.4 points apart on average over 3 shared benchmarks.
- Widest single gap
- LMArena Text: Chinese: Llama 4 Maverick 1277 against Qwen3 235B A22B Thinking 2507 1481, 34.0 points apart on a 0 to 100 scale.
- Output price
- Llama 4 Maverick $1.00 against Qwen3 235B A22B Thinking 2507 $3.50 per million output tokens, 3.5x apart.
- Context window
- Llama 4 Maverick takes 524K 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 | Llama 4 Maverick | Bars | Qwen3 235B A22B Thinking 2507 | Ties |
|---|---|---|---|---|
| Coding | 0 | 4 | ||
| Intelligence | 0 | 1 | ||
| Agentic work | 0 | 1 | ||
| Reasoning | 0 | 3 | ||
| Math | 0 | 2 | ||
| Knowledge | 0 | 4 | ||
| Web design | 0 | 6 | ||
| General | 0 | 20 | ||
| All shared boards | 0 | 41 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index04
| Benchmark | Llama 4 Maverick | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| Aider polyglot | 15.6% | Not measured | Only Llama 4 Maverick measured |
| SWE-bench Verified | 21.0% | Not measured | Only Llama 4 Maverick measured |
| AA Coding Index | 16.3 | 22.1 | Qwen3 235B A22B Thinking 2507 |
| LiveCodeBench (Vals AI) | 47.3% | Not measured | Only Llama 4 Maverick measured |
| LiveCodeBench (Artificial Analysis run) | 39.7% | Not measured | Only Llama 4 Maverick measured |
| SciCode | 31.7% | Not measured | Only Llama 4 Maverick measured |
| Terminal-Bench Hard (Artificial Analysis run) | 6.8% | Not measured | Only Llama 4 Maverick measured |
| WeirdML | 24.5% | 41.0% | Qwen3 235B A22B Thinking 2507 |
| ALE-Bench | 173.0 | Not measured | Only Llama 4 Maverick measured |
| LMArena Text: coding | 1301 | 1424 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: software and IT | 1299 | 1431 | Qwen3 235B A22B Thinking 2507 |
| Neither model has a result on the other 32 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Llama 4 Maverick | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| AA Intelligence Index | 9.3 | 12.7 | Qwen3 235B A22B Thinking 2507 |
| Epoch Capabilities Index | 132.2 | Not measured | Only Llama 4 Maverick measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Llama 4 Maverick | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| AA Agentic Index | 0.6 | 1.3 | Qwen3 235B A22B Thinking 2507 |
| τ²-Bench Telecom | 17.8% | Not measured | Only Llama 4 Maverick measured |
| Vending-Bench 2 | Not measured | -$11 | Only Qwen3 235B A22B Thinking 2507 measured |
| Neither model has a result on the other 39 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index03
| Benchmark | Llama 4 Maverick | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| GPQA Diamond | 69.4% | Not measured | Only Llama 4 Maverick measured |
| MMMU Pro (Vals AI) | 71.7% | Not measured | Only Llama 4 Maverick measured |
| GPQA Diamond (Artificial Analysis run) | 67.1% | Not measured | Only Llama 4 Maverick measured |
| Humanity's Last Exam (Artificial Analysis run) | 4.9% | Not measured | Only Llama 4 Maverick measured |
| GPQA Diamond (Epoch AI run) | Not measured | 80.1% | Only Qwen3 235B A22B Thinking 2507 measured |
| ARC-AGI | 4.4% | Not measured | Only Llama 4 Maverick measured |
| ARC-AGI-2 | 0.0% | Not measured | Only Llama 4 Maverick measured |
| SimpleBench | 27.7% | Not measured | Only Llama 4 Maverick measured |
| Mystery game puzzles | Not measured | 9.0% | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: hard prompts | 1281 | 1416 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: expert | 1259 | 1463 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: hard prompts in English | 1295 | 1432 | Qwen3 235B A22B Thinking 2507 |
| LMArena Vision: diagrams | 1134 | Not measured | Only Llama 4 Maverick measured |
| LMArena Vision: homework | 1164 | Not measured | Only Llama 4 Maverick measured |
| Kagi LLM Benchmark | 55.9% | Not measured | Only Llama 4 Maverick measured |
| Neither model has a result on the other 8 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Llama 4 Maverick | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| AIME | 25.2% | Not measured | Only Llama 4 Maverick measured |
| MATH 500 (Vals AI) | 85.2% | Not measured | Only Llama 4 Maverick measured |
| MGSM (Vals AI) | 92.4% | Not measured | Only Llama 4 Maverick measured |
| AA Math Index | 19.3 | Not measured | Only Llama 4 Maverick measured |
| AIME 2025 (Artificial Analysis run) | 19.3% | Not measured | Only Llama 4 Maverick measured |
| OTIS Mock AIME | Not measured | 86.7% | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: math | 1299 | 1412 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: mathematical work | 1295 | 1435 | Qwen3 235B A22B Thinking 2507 |
| Neither model has a result on the other 6 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index04
| Benchmark | Llama 4 Maverick | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| MMLU Pro | 79.4% | Not measured | Only Llama 4 Maverick measured |
| LegalBench (Vals AI) | 77.8% | Not measured | Only Llama 4 Maverick measured |
| CorpFin (Vals AI) | 49.7% | Not measured | Only Llama 4 Maverick measured |
| TaxEval (Vals AI) | 66.6% | Not measured | Only Llama 4 Maverick measured |
| MortgageTax (Vals AI) | 58.5% | Not measured | Only Llama 4 Maverick measured |
| MedQA (Vals AI) | 43.3% | Not measured | Only Llama 4 Maverick measured |
| MedCode (Vals AI) | 36.5% | Not measured | Only Llama 4 Maverick measured |
| MMLU Pro (Artificial Analysis run) | 80.9% | Not measured | Only Llama 4 Maverick measured |
| SimpleQA Verified | Not measured | 40.4% | Only Qwen3 235B A22B Thinking 2507 measured |
| GeoBench | 52.0% | Not measured | Only Llama 4 Maverick measured |
| LMArena Text: science | 1288 | 1433 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: medicine and healthcare | 1275 | 1456 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: legal and government | 1286 | 1415 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: business and finance | 1260 | 1413 | Qwen3 235B A22B Thinking 2507 |
| Neither model has a result on the other 6 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index06
| Benchmark | Llama 4 Maverick | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| Design Arena: websites | 871 | 1052 | Qwen3 235B A22B Thinking 2507 |
| Design Arena: UI components | 897 | 942 | Qwen3 235B A22B Thinking 2507 |
| Design Arena: game dev | 837 | 956 | Qwen3 235B A22B Thinking 2507 |
| Design Arena: data viz | 882 | 950 | Qwen3 235B A22B Thinking 2507 |
| Design Arena: 3D | 900 | 997 | Qwen3 235B A22B Thinking 2507 |
| Design Arena: all code categories | 879 | 1032 | Qwen3 235B A22B Thinking 2507 |
| Neither model has a result on the other 3 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index020
| Benchmark | Llama 4 Maverick | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| SAGE (Vals AI) | 35.7% | Not measured | Only Llama 4 Maverick measured |
| MedScribe (Vals AI) | 54.2% | Not measured | Only Llama 4 Maverick measured |
| IFBench | 43.0% | Not measured | Only Llama 4 Maverick measured |
| Long context reasoning (AA-LCR) | 50.0% | Not measured | Only Llama 4 Maverick measured |
| DTBench | 61.9% | 80.3% | Qwen3 235B A22B Thinking 2507 |
| LMCA | 15.9% | 29.3% | Qwen3 235B A22B Thinking 2507 |
| Fiction.LiveBench | 46.2% | 75.0% | Qwen3 235B A22B Thinking 2507 |
| Creative writing (Lech Mazur) | 62.0% | 82.4% | Qwen3 235B A22B Thinking 2507 |
| LMArena Text | 1287 | 1415 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: instruction following | 1267 | 1386 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: creative writing | 1266 | 1384 | Qwen3 235B A22B Thinking 2507 |
| LMArena Vision | 1142 | Not measured | Only Llama 4 Maverick measured |
| LMArena Text: English | 1303 | 1429 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: non-English | 1269 | 1397 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: Chinese | 1277 | 1481 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: German | 1291 | 1408 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: Japanese | 1206 | 1386 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: Korean | 1203 | 1375 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: French | 1259 | Not measured | Only Llama 4 Maverick measured |
| LMArena Text: Spanish | 1293 | 1389 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: Russian | 1286 | 1401 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: Polish | 1275 | 1390 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: multi-turn | 1289 | 1406 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: longer queries | 1280 | 1400 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: writing and language | 1277 | 1388 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: entertainment and media | 1250 | 1370 | Qwen3 235B A22B Thinking 2507 |
| LMArena Vision: English | 1150 | Not measured | Only Llama 4 Maverick measured |
| LMArena Vision: Chinese | 1141 | Not measured | Only Llama 4 Maverick measured |
| LMArena Vision: reading text | 1143 | Not measured | Only Llama 4 Maverick measured |
| LMArena Vision: humor | 1100 | Not measured | Only Llama 4 Maverick measured |
| LMArena Vision: creative writing | 1099 | Not measured | Only Llama 4 Maverick measured |
| Neither model has a result on the other 17 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Llama 4 Maverick | 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
Llama 4 Maverick costs less
Repository review
50K input + 3K output tokens
Llama 4 Maverick costs less
Agent loop
200K input + 10K output tokens
Llama 4 Maverick 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 | Meta Llama 4 Maverick | Qwen3 235B A22B Thinking 2507 |
|---|---|---|
| Maker | Meta Llama | Qwen |
| Weights | Open weights | Open weights |
| Released | April 5, 2025 | July 25, 2025 |
| Context window | 524K | 128K |
| Input $/Mtok | $0.35 | $0.45 |
| Output $/Mtok | $1.00 | $3.50 |
| Coding index | Not benchmarked | Not benchmarked |
| Independent measurements behind it | 0 | 0 |
| Cost per solved task | $1.491 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Llama 4 Maverick better than Qwen3 235B A22B Thinking 2507 for coding?
Neither model has a coding score yet, so the two cannot be ranked on coding. Pick Llama 4 Maverick for output tokens 3.5x cheaper ($1.00 vs $3.50 per million) and a larger context window (524K against 128K tokens).
Which is cheaper, Llama 4 Maverick or Qwen3 235B A22B Thinking 2507?
Llama 4 Maverick costs $0.35 per million input tokens and $1.00 per million output tokens; Qwen3 235B A22B Thinking 2507 costs $0.45 and $3.50. Cost per solved task is known for Llama 4 Maverick only.
Which has the bigger context window, Llama 4 Maverick or Qwen3 235B A22B Thinking 2507?
Llama 4 Maverick, with 524K tokens against 128K.
How many benchmarks is the Llama 4 Maverick vs Qwen3 235B A22B Thinking 2507 comparison based on?
41 benchmarks have a result for both models as of October 22, 2026. Llama 4 Maverick's coding index rests on 0 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.