Qwen3 235B A22B Thinking 2507 vs Qwen3 32B
Qwen3 235B A22B Thinking 2507 leads Qwen3 32B on the Artificial Analysis Intelligence Index, 12.7 to 7.3, and Qwen3 32B costs 8.8x less per output token ($0.40 vs $3.50 per million). Updated October 22, 2026; 29 benchmarks have a result for both.
Qwen3 235B A22B Thinking 2507 or Qwen3 32B: which to pick
Neither model has a coding score yet, so the two cannot be ranked on coding.
Compare them on price, context and the boards below.
- Shared results
- 29
- Qwen3 235B A22B Thinking 2507 only
- 17
- Qwen3 32B only
- 16
- 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
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 7.3.
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
No call
Only Qwen3 235B A22B Thinking 2507 has a Design Arena: websites result, 1052.
Directional only
Long documents
Listed context window
Qwen3 32B
Qwen3 32B takes 131K tokens against 128K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Qwen3 32B
Qwen3 235B A22B Thinking 2507 $0.0022, Qwen3 32B $0.0003 per turn.
List prices
At a glance
- Coding index
- Qwen3 235B A22B Thinking 2507 not benchmarked, Qwen3 32B not benchmarked.
- Widest category
- Reasoning, 14.6 points apart on average over 4 shared benchmarks.
- Widest single gap
- LMArena Text: Chinese: Qwen3 235B A22B Thinking 2507 1481 against Qwen3 32B 1357, 20.7 points apart on a 0 to 100 scale.
- Output price
- Qwen3 235B A22B Thinking 2507 $3.50 against Qwen3 32B $0.40 per million output tokens, 8.8x apart.
- Context window
- Qwen3 235B A22B Thinking 2507 takes 128K tokens, Qwen3 32B 131K.
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 235B A22B Thinking 2507 | Bars | Qwen3 32B | Ties |
|---|---|---|---|---|
| Coding | 2 | 0 | ||
| Intelligence | 1 | 0 | ||
| Reasoning | 4 | 0 | ||
| Math | 3 | 0 | ||
| Knowledge | 4 | 0 | ||
| General | 15 | 0 | ||
| All shared boards | 29 | 0 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index20
| Benchmark | Qwen3 235B A22B Thinking 2507 | Qwen3 32B | Leader |
|---|---|---|---|
| Aider polyglot | Not measured | 40.0% | Only Qwen3 32B measured |
| AA Coding Index | 22.1 | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| LiveCodeBench (Artificial Analysis run) | Not measured | 54.6% | Only Qwen3 32B measured |
| SciCode | Not measured | 36.0% | Only Qwen3 32B measured |
| Terminal-Bench Hard (Artificial Analysis run) | Not measured | 3.0% | Only Qwen3 32B measured |
| WeirdML | 41.0% | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: coding | 1424 | 1357 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: software and IT | 1431 | 1366 | Qwen3 235B A22B Thinking 2507 |
| Neither model has a result on the other 35 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | Qwen3 235B A22B Thinking 2507 | Qwen3 32B | Leader |
|---|---|---|---|
| AA Intelligence Index | 12.7 | 7.3 | Qwen3 235B A22B Thinking 2507 |
| Epoch Capabilities Index | Not measured | 138.4 | Only Qwen3 32B measured |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index5 boards, none shared
| Benchmark | Qwen3 235B A22B Thinking 2507 | Qwen3 32B | Leader |
|---|---|---|---|
| AA Agentic Index | 1.3 | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| τ²-Bench Telecom | Not measured | 29.8% | Only Qwen3 32B measured |
| Vending-Bench 2 | -$11 | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| Berkeley Function Calling Leaderboard (BFCL) | Not measured | 48.7% | Only Qwen3 32B measured |
| BrowseComp-Plus | Not measured | 10.7% | Only Qwen3 32B measured |
| Neither model has a result on the other 37 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index40
| Benchmark | Qwen3 235B A22B Thinking 2507 | Qwen3 32B | Leader |
|---|---|---|---|
| GPQA Diamond (Artificial Analysis run) | Not measured | 66.8% | Only Qwen3 32B measured |
| Humanity's Last Exam (Artificial Analysis run) | Not measured | 7.4% | Only Qwen3 32B measured |
| GPQA Diamond (Epoch AI run) | 80.1% | 65.7% | Qwen3 235B A22B Thinking 2507 |
| Mystery game puzzles | 9.0% | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: hard prompts | 1416 | 1334 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: expert | 1463 | 1362 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: hard prompts in English | 1432 | 1350 | Qwen3 235B A22B Thinking 2507 |
| Kagi LLM Benchmark | Not measured | 48.7% | Only Qwen3 32B measured |
| Neither model has a result on the other 15 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index30
| Benchmark | Qwen3 235B A22B Thinking 2507 | Qwen3 32B | Leader |
|---|---|---|---|
| AA Math Index | Not measured | 73.0 | Only Qwen3 32B measured |
| AIME 2025 (Artificial Analysis run) | Not measured | 73.0% | Only Qwen3 32B measured |
| OTIS Mock AIME | 86.7% | 66.9% | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: math | 1412 | 1399 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: mathematical work | 1435 | 1420 | Qwen3 235B A22B Thinking 2507 |
| Neither model has a result on the other 9 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index40
| Benchmark | Qwen3 235B A22B Thinking 2507 | Qwen3 32B | Leader |
|---|---|---|---|
| MMLU Pro (Artificial Analysis run) | Not measured | 79.8% | Only Qwen3 32B measured |
| SimpleQA Verified | 40.4% | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: science | 1433 | 1372 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: medicine and healthcare | 1456 | 1368 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: legal and government | 1415 | 1324 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: business and finance | 1413 | 1318 | Qwen3 235B A22B Thinking 2507 |
| Neither model has a result on the other 14 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index6 boards, none shared
| Benchmark | Qwen3 235B A22B Thinking 2507 | Qwen3 32B | Leader |
|---|---|---|---|
| Design Arena: websites | 1052 | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| Design Arena: UI components | 942 | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| Design Arena: game dev | 956 | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| Design Arena: data viz | 950 | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| Design Arena: 3D | 997 | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| Design Arena: all code categories | 1032 | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| Neither model has a result on the other 3 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index150
| Benchmark | Qwen3 235B A22B Thinking 2507 | Qwen3 32B | Leader |
|---|---|---|---|
| IFBench | Not measured | 36.3% | Only Qwen3 32B measured |
| Long context reasoning (AA-LCR) | Not measured | 0.0% | Only Qwen3 32B measured |
| DTBench | 80.3% | 67.5% | Qwen3 235B A22B Thinking 2507 |
| LMCA | 29.3% | 17.3% | Qwen3 235B A22B Thinking 2507 |
| Fiction.LiveBench | 75.0% | 74.2% | Qwen3 235B A22B Thinking 2507 |
| Creative writing (Lech Mazur) | 82.4% | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text | 1415 | 1340 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: instruction following | 1386 | 1305 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: creative writing | 1384 | 1297 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: English | 1429 | 1359 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: non-English | 1397 | 1317 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: Chinese | 1481 | 1357 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: German | 1408 | 1341 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: Japanese | 1386 | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: Korean | 1375 | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: Spanish | 1389 | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: Russian | 1401 | 1311 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: Polish | 1390 | Not measured | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: multi-turn | 1406 | 1331 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: longer queries | 1400 | 1327 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: writing and language | 1388 | 1298 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: entertainment and media | 1370 | 1279 | Qwen3 235B A22B Thinking 2507 |
| Neither model has a result on the other 26 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Qwen3 235B A22B Thinking 2507 | Qwen3 32B | 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 32B costs less
Repository review
50K input + 3K output tokens
Qwen3 32B costs less
Agent loop
200K input + 10K output tokens
Qwen3 32B 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 235B A22B Thinking 2507 | Qwen3 32B |
|---|---|---|
| Maker | Qwen | Qwen |
| Weights | Open weights | Open weights |
| Released | July 25, 2025 | April 28, 2025 |
| Context window | 128K | 131K |
| Input $/Mtok | $0.45 | $0.14 |
| Output $/Mtok | $3.50 | $0.40 |
| Coding index | Not benchmarked | Not benchmarked |
| Independent measurements behind it | 0 | 0 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Qwen3 235B A22B Thinking 2507 better than Qwen3 32B for coding?
Neither model has a coding score yet, so the two cannot be ranked on coding. Compare them on price, context and the boards below.
Which is cheaper, Qwen3 235B A22B Thinking 2507 or Qwen3 32B?
Qwen3 235B A22B Thinking 2507 costs $0.45 per million input tokens and $3.50 per million output tokens; Qwen3 32B costs $0.14 and $0.40. Cost per solved task is known for neither of them.
Which has the bigger context window, Qwen3 235B A22B Thinking 2507 or Qwen3 32B?
Qwen3 32B, with 131K tokens against 128K.
How many benchmarks is the Qwen3 235B A22B Thinking 2507 vs Qwen3 32B comparison based on?
29 benchmarks have a result for both models as of October 22, 2026. Qwen3 235B A22B Thinking 2507's coding index rests on 0 independent benchmarks and Qwen3 32B's on 0. Where only one model has a result, the page says so instead of naming a winner.