GPT-4o vs Qwen3 235B A22B Thinking 2507
Qwen3 235B A22B Thinking 2507 costs 2.9x less per output token than GPT-4o ($3.50 vs $10.00 per million). Updated October 22, 2026; 8 benchmarks have a result for both.
GPT-4o 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 GPT-4o for work measured by Design Arena: websites or Design Arena: UI components, where it leads.
- Shared results
- 8
- GPT-4o only
- 19
- Qwen3 235B A22B Thinking 2507 only
- 38
- Like-for-like categories
- 2 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
No call
Only Qwen3 235B A22B Thinking 2507 has a AA Intelligence Index result, 12.7.
Directional only
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
Qwen3 235B A22B Thinking 2507
Qwen3 235B A22B Thinking 2507 leads, 1052 against 831.
Same board, both measured
Long documents
Listed context window
No call
Both take 128K tokens.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Qwen3 235B A22B Thinking 2507
GPT-4o $0.0075, Qwen3 235B A22B Thinking 2507 $0.0022 per turn.
List prices
At a glance
- Coding index
- GPT-4o not benchmarked, Qwen3 235B A22B Thinking 2507 not benchmarked.
- Widest category
- Web design, 15.1 points apart on average over 6 shared benchmarks.
- Widest single gap
- Design Arena: websites: GPT-4o 831 against Qwen3 235B A22B Thinking 2507 1052, 27.6 points apart on a 0 to 100 scale.
- Output price
- GPT-4o $10.00 against Qwen3 235B A22B Thinking 2507 $3.50 per million output tokens, 2.9x apart.
- Context window
- Both take 128K tokens.
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-4o | Bars | Qwen3 235B A22B Thinking 2507 | Ties |
|---|---|---|---|---|
| Web design | 0 | 6 | ||
| General | 0 | 2 | ||
| All shared boards | 0 | 8 |
CodingDirectional onlySix of these boards feed the Vibe Coding Index11 boards, none shared
| Benchmark | GPT-4o | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| SWE-bench Verified | 38.8% | Not measured | Only GPT-4o measured |
| SWE-bench Lite | 38.0% | Not measured | Only GPT-4o measured |
| SWE-bench (full test set) | 18.8% | Not measured | Only GPT-4o measured |
| SWE-bench Multimodal | 30.4% | Not measured | Only GPT-4o measured |
| AA Coding Index | Not measured | 22.1 | Only Qwen3 235B A22B Thinking 2507 measured |
| LiveCodeBench (Artificial Analysis run) | 30.9% | Not measured | Only GPT-4o measured |
| Terminal-Bench Hard (Artificial Analysis run) | 8.3% | Not measured | Only GPT-4o measured |
| WeirdML | Not measured | 41.0% | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: coding | Not measured | 1424 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: software and IT | Not measured | 1431 | Only Qwen3 235B A22B Thinking 2507 measured |
| SciCode | 1.5% | Not measured | Only GPT-4o measured |
| Neither model has a result on the other 28 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index1 board, none shared
| Benchmark | GPT-4o | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| AA Intelligence Index | Not measured | 12.7 | Only Qwen3 235B A22B Thinking 2507 measured |
| Neither model has a result on the other 1 board here | |||
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index4 boards, none shared
| Benchmark | GPT-4o | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| AA Agentic Index | Not measured | 1.3 | Only Qwen3 235B A22B Thinking 2507 measured |
| τ²-Bench Telecom | 25.1% | Not measured | Only GPT-4o measured |
| GDPval | 9.9% | Not measured | Only GPT-4o measured |
| Vending-Bench 2 | Not measured | -$11 | Only Qwen3 235B A22B Thinking 2507 measured |
| Neither model has a result on the other 37 boards here | |||
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index9 boards, none shared
| Benchmark | GPT-4o | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| GPQA Diamond (Artificial Analysis run) | 54.3% | Not measured | Only GPT-4o measured |
| Humanity's Last Exam (Artificial Analysis run) | 2.4% | Not measured | Only GPT-4o measured |
| GPQA Diamond (Epoch AI run) | Not measured | 80.1% | Only Qwen3 235B A22B Thinking 2507 measured |
| ARC-AGI | 4.5% | Not measured | Only GPT-4o measured |
| ARC-AGI-2 | 0.0% | Not measured | Only GPT-4o measured |
| Mystery game puzzles | Not measured | 9.0% | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: hard prompts | Not measured | 1416 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: expert | Not measured | 1463 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: hard prompts in English | Not measured | 1432 | Only Qwen3 235B A22B Thinking 2507 measured |
| Neither model has a result on the other 13 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index5 boards, none shared
| Benchmark | GPT-4o | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| AA Math Index | 6.0 | Not measured | Only GPT-4o measured |
| AIME 2025 (Artificial Analysis run) | 6.0% | Not measured | Only GPT-4o measured |
| OTIS Mock AIME | Not measured | 86.7% | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: math | Not measured | 1412 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: mathematical work | Not measured | 1435 | Only Qwen3 235B A22B Thinking 2507 measured |
| Neither model has a result on the other 8 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index7 boards, none shared
| Benchmark | GPT-4o | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| MMLU Pro (Artificial Analysis run) | 74.8% | Not measured | Only GPT-4o measured |
| SimpleQA Verified | Not measured | 40.4% | Only Qwen3 235B A22B Thinking 2507 measured |
| MMLU | 88.1% | Not measured | Only GPT-4o measured |
| LMArena Text: science | Not measured | 1433 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: medicine and healthcare | Not measured | 1456 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: legal and government | Not measured | 1415 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: business and finance | Not measured | 1413 | Only Qwen3 235B A22B Thinking 2507 measured |
| Neither model has a result on the other 11 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index06
| Benchmark | GPT-4o | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| Design Arena: websites | 831 | 1052 | Qwen3 235B A22B Thinking 2507 |
| Design Arena: UI components | 882 | 942 | Qwen3 235B A22B Thinking 2507 |
| Design Arena: game dev | 906 | 956 | Qwen3 235B A22B Thinking 2507 |
| Design Arena: data viz | 858 | 950 | Qwen3 235B A22B Thinking 2507 |
| Design Arena: 3D | 868 | 997 | Qwen3 235B A22B Thinking 2507 |
| Design Arena: all code categories | 860 | 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 index02
| Benchmark | GPT-4o | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| IFBench | 34.3% | Not measured | Only GPT-4o measured |
| Long context reasoning (AA-LCR) | 49.3% | Not measured | Only GPT-4o measured |
| DTBench | 64.5% | 80.3% | Qwen3 235B A22B Thinking 2507 |
| LMCA | 16.6% | 29.3% | Qwen3 235B A22B Thinking 2507 |
| 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 |
| LMArena Text | Not measured | 1415 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: instruction following | Not measured | 1386 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: creative writing | Not measured | 1384 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: English | Not measured | 1429 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: non-English | Not measured | 1397 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: Chinese | Not measured | 1481 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: German | Not measured | 1408 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: Japanese | Not measured | 1386 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: Korean | Not measured | 1375 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: Spanish | Not measured | 1389 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: Russian | Not measured | 1401 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: Polish | Not measured | 1390 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: multi-turn | Not measured | 1406 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: longer queries | Not measured | 1400 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: writing and language | Not measured | 1388 | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: entertainment and media | Not measured | 1370 | Only Qwen3 235B A22B Thinking 2507 measured |
| Neither model has a result on the other 24 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-4o | 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
Qwen3 235B A22B Thinking 2507 costs less
Repository review
50K input + 3K output tokens
Qwen3 235B A22B Thinking 2507 costs less
Agent loop
200K input + 10K output tokens
Qwen3 235B A22B Thinking 2507 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-4o | Qwen3 235B A22B Thinking 2507 |
|---|---|---|
| Maker | OpenAI | Qwen |
| Weights | No public weights listed | Open weights |
| Released | May 13, 2024 | July 25, 2025 |
| Context window | 128K | 128K |
| Input $/Mtok | $2.50 | $0.45 |
| Output $/Mtok | $10.00 | $3.50 |
| 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 GPT-4o 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 GPT-4o for work measured by Design Arena: websites or Design Arena: UI components, where it leads.
Which is cheaper, GPT-4o or Qwen3 235B A22B Thinking 2507?
GPT-4o costs $2.50 per million input tokens and $10.00 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-4o or Qwen3 235B A22B Thinking 2507?
Neither: both take 128K tokens.
How many benchmarks is the GPT-4o vs Qwen3 235B A22B Thinking 2507 comparison based on?
8 benchmarks have a result for both models as of October 22, 2026. GPT-4o'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.