GPT-5.2-Codex vs Qwen3.7 Max
Qwen3.7 Max leads GPT-5.2-Codex on coding, 54.1 to 45.4 on the Vibe Coding Index, and Qwen3.7 Max costs 3.2x less per output token ($4.425 vs $14.00 per million). Updated October 21, 2026; 19 benchmarks have a result for both.
GPT-5.2-Codex or Qwen3.7 Max: which to pick
Pick Qwen3.7 Max for coding: it scores 54.1 on the Vibe Coding Index against 45.4 for GPT-5.2-Codex.
Pick GPT-5.2-Codex for a lower cost per solved task ($0.43 against $4.69) and work measured by LiveCodeBench (Vals AI) or SWE-bench Verified (Vals AI), where it leads.
- Shared results
- 19
- GPT-5.2-Codex only
- 3
- Qwen3.7 Max only
- 40
- Like-for-like categories
- 5 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
Qwen3.7 Max
Qwen3.7 Max scores 54.1 against 45.4.
3 benchmarks behind each score
Intelligence
AA Intelligence Index
Qwen3.7 Max
Qwen3.7 Max leads, 29.5 against 28.5.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only Qwen3.7 Max has a AA Agentic Index result, 22.5.
Directional only
Reasoning
GPQA Diamond
No call
Only Qwen3.7 Max has a GPQA Diamond result, 90.2%.
Directional only
Knowledge
MMLU Pro
No call
Only Qwen3.7 Max has a MMLU Pro result, 89.3%.
Directional only
Web design
Design Arena: websites
No call
Only Qwen3.7 Max has a Design Arena: websites result, 1274.
Directional only
Long documents
Listed context window
Qwen3.7 Max
Qwen3.7 Max takes 1M tokens against 400K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Qwen3.7 Max
GPT-5.2-Codex $0.0088, Qwen3.7 Max $0.0037 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
GPT-5.2-Codex
GPT-5.2-Codex $0.43, Qwen3.7 Max $4.69.
Published attempt cost and pass rate
At a glance
- Coding index
- GPT-5.2-Codex 45.4 from 3 benchmarks, Qwen3.7 Max 54.1 from 3 benchmarks.
- Widest category
- Coding, 14.3 points apart on average over 8 shared benchmarks.
- Widest single gap
- LMArena WebDev: React: GPT-5.2-Codex 1329 against Qwen3.7 Max 1514, 30.7 points apart on a 0 to 100 scale.
- Output price
- GPT-5.2-Codex $14.00 against Qwen3.7 Max $4.425 per million output tokens, 3.2x apart.
- Context window
- GPT-5.2-Codex takes 400K tokens, Qwen3.7 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-5.2-Codex | Bars | Qwen3.7 Max | Ties |
|---|---|---|---|---|
| Coding | 3 | 5 | ||
| Intelligence | 0 | 1 | ||
| Agentic work | 1 | 5 | ||
| Reasoning | 0 | 2 | ||
| General | 1 | 1 | ||
| All shared boards | 5 | 14 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index35
| Benchmark | GPT-5.2-Codex | Qwen3.7 Max | Leader |
|---|---|---|---|
| SWE-bench Verified | 72.8% | Not measured | Only GPT-5.2-Codex measured |
| SWE-bench Multilingual | 66.3% | Not measured | Only GPT-5.2-Codex measured |
| LMArena WebDev | 1339 | 1515 | Qwen3.7 Max |
| AA Coding Index | Not measured | 66.0 | Only Qwen3.7 Max measured |
| LiveCodeBench (Vals AI) | 88.0% | 87.1% | GPT-5.2-Codex |
| SWE-bench Verified (Vals AI) | 72.4% | 68.8% | GPT-5.2-Codex |
| Vibe Code Bench (Vals AI) | 37.9% | 47.7% | Qwen3.7 Max |
| Code migration (Vals AI) | Not measured | 13.1% | Only Qwen3.7 Max measured |
| Terminal-Bench 2.1 (Vals AI) | Not measured | 61.0% | Only Qwen3.7 Max measured |
| SciCode | Not measured | 49.5% | Only Qwen3.7 Max measured |
| Terminal-Bench Hard (Artificial Analysis run) | 37.1% | 50.8% | Qwen3.7 Max |
| Terminal-Bench | 66.5% | Not measured | Only GPT-5.2-Codex measured |
| ALE-Bench | 1299.9 | 1189.4 | GPT-5.2-Codex |
| LMArena WebDev: React | 1329 | 1514 | Qwen3.7 Max |
| LMArena WebDev: HTML | 1348 | 1512 | Qwen3.7 Max |
| Neither model has a result on the other 24 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | GPT-5.2-Codex | Qwen3.7 Max | Leader |
|---|---|---|---|
| AA Intelligence Index | 28.5 | 29.5 | Qwen3.7 Max |
| Epoch Capabilities Index | Not measured | 154.0 | Only Qwen3.7 Max measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index15
| Benchmark | GPT-5.2-Codex | Qwen3.7 Max | Leader |
|---|---|---|---|
| AA Agentic Index | Not measured | 22.5 | Only Qwen3.7 Max measured |
| Design Arena: agents building web apps | 1051 | 1208 | Qwen3.7 Max |
| Design Arena: agents building full-stack apps | 993 | 1191 | Qwen3.7 Max |
| Design Arena: agents building mobile apps | 1090 | 1155 | Qwen3.7 Max |
| Design Arena: agents building Android apps | 1176 | 1159 | GPT-5.2-Codex |
| Design Arena: agents building Godot games | 1142 | 1187 | Qwen3.7 Max |
| Design Arena: agents building games | Not measured | 1161 | Only Qwen3.7 Max measured |
| Design Arena: HTML slides | Not measured | 1158 | Only Qwen3.7 Max measured |
| Design Arena: PowerPoint slides | Not measured | 1197 | Only Qwen3.7 Max measured |
| Finance Agent (Vals AI) | Not measured | 47.8% | Only Qwen3.7 Max measured |
| Tax Agent Bench (Vals AI) | Not measured | 15.8% | Only Qwen3.7 Max measured |
| Legal Research Bench (Vals AI) | Not measured | 25.5% | Only Qwen3.7 Max measured |
| Excel Modeling Benchmark (Vals AI) | Not measured | 57.0% | Only Qwen3.7 Max measured |
| τ²-Bench Telecom | 92.1% | 94.7% | Qwen3.7 Max |
| Neither model has a result on the other 27 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | GPT-5.2-Codex | Qwen3.7 Max | Leader |
|---|---|---|---|
| GPQA Diamond | Not measured | 90.2% | Only Qwen3.7 Max measured |
| GPQA Diamond (Artificial Analysis run) | 89.9% | 92.3% | Qwen3.7 Max |
| Humanity's Last Exam (Artificial Analysis run) | 35.7% | 40.5% | Qwen3.7 Max |
| GPQA Diamond (Epoch AI run) | Not measured | 90.9% | Only Qwen3.7 Max measured |
| SimpleBench | Not measured | 70.4% | Only Qwen3.7 Max measured |
| Chess puzzles | Not measured | 26.0% | Only Qwen3.7 Max measured |
| Mystery game puzzles | Not measured | 32.0% | Only Qwen3.7 Max measured |
| EBR-bench | Not measured | 9.5% | Only Qwen3.7 Max measured |
| Neither model has a result on the other 14 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index4 boards, none shared
| Benchmark | GPT-5.2-Codex | Qwen3.7 Max | Leader |
|---|---|---|---|
| OTIS Mock AIME | Not measured | 95.6% | Only Qwen3.7 Max measured |
| FrontierMath, tiers 1 to 3 | Not measured | 64.6% | Only Qwen3.7 Max measured |
| FrontierMath, tier 4 | Not measured | 34.1% | Only Qwen3.7 Max measured |
| ProofBench | Not measured | 26.0% | Only Qwen3.7 Max measured |
| Neither model has a result on the other 9 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index6 boards, none shared
| Benchmark | GPT-5.2-Codex | Qwen3.7 Max | Leader |
|---|---|---|---|
| MMLU Pro | Not measured | 89.3% | Only Qwen3.7 Max measured |
| LegalBench (Vals AI) | Not measured | 84.9% | Only Qwen3.7 Max measured |
| CorpFin (Vals AI) | Not measured | 63.7% | Only Qwen3.7 Max measured |
| TaxEval (Vals AI) | Not measured | 75.3% | Only Qwen3.7 Max measured |
| MedCode (Vals AI) | Not measured | 38.8% | Only Qwen3.7 Max measured |
| SimpleQA Verified | Not measured | 55.8% | Only Qwen3.7 Max measured |
| Neither model has a result on the other 12 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
| Benchmark | GPT-5.2-Codex | Qwen3.7 Max | Leader |
|---|---|---|---|
| Design Arena: websites | Not measured | 1274 | Only Qwen3.7 Max measured |
| Design Arena: UI components | Not measured | 1270 | Only Qwen3.7 Max measured |
| Design Arena: game dev | Not measured | 1265 | Only Qwen3.7 Max measured |
| Design Arena: data viz | Not measured | 1275 | Only Qwen3.7 Max measured |
| Design Arena: 3D | Not measured | 1266 | Only Qwen3.7 Max measured |
| Design Arena: all code categories | Not measured | 1275 | Only Qwen3.7 Max measured |
| Design Arena: SVG | Not measured | 1217 | Only Qwen3.7 Max measured |
| Design Arena: ASCII art | Not measured | 1211 | Only Qwen3.7 Max measured |
| Neither model has a result on the other 1 board here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index11
| Benchmark | GPT-5.2-Codex | Qwen3.7 Max | Leader |
|---|---|---|---|
| MedScribe (Vals AI) | Not measured | 79.4% | Only Qwen3.7 Max measured |
| IFBench | 77.6% | 80.5% | Qwen3.7 Max |
| Long context reasoning (AA-LCR) | 82.3% | 79.0% | GPT-5.2-Codex |
| DTBench | Not measured | 92.3% | Only Qwen3.7 Max measured |
| LMCA | Not measured | 44.0% | Only Qwen3.7 Max measured |
| Neither model has a result on the other 41 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-5.2-Codex | Qwen3.7 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
Qwen3.7 Max costs less
Repository review
50K input + 3K output tokens
Qwen3.7 Max costs less
Agent loop
200K input + 10K output tokens
Qwen3.7 Max 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-5.2-Codex | Qwen3.7 Max |
|---|---|---|
| Maker | OpenAI | Qwen |
| Weights | No public weights listed | No public weights listed |
| Released | January 14, 2026 | May 21, 2026 |
| Context window | 400K | 1M |
| Input $/Mtok | $1.75 | $1.475 |
| Output $/Mtok | $14.00 | $4.425 |
| Coding index | 45.4 | 54.1 |
| Independent measurements behind it | 3 | 3 |
| Cost per solved task | $0.43 | $4.69 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GPT-5.2-Codex better than Qwen3.7 Max for coding?
Pick Qwen3.7 Max for coding: it scores 54.1 on the Vibe Coding Index against 45.4 for GPT-5.2-Codex. Pick GPT-5.2-Codex for a lower cost per solved task ($0.43 against $4.69) and work measured by LiveCodeBench (Vals AI) or SWE-bench Verified (Vals AI), where it leads.
Which is cheaper, GPT-5.2-Codex or Qwen3.7 Max?
GPT-5.2-Codex costs $1.75 per million input tokens and $14.00 per million output tokens; Qwen3.7 Max costs $1.475 and $4.425. Per solved SWE-bench Verified task, GPT-5.2-Codex costs $0.43 and Qwen3.7 Max $4.69.
Which has the bigger context window, GPT-5.2-Codex or Qwen3.7 Max?
Qwen3.7 Max, with 1M tokens against 400K.
How many benchmarks is the GPT-5.2-Codex vs Qwen3.7 Max comparison based on?
19 benchmarks have a result for both models as of October 21, 2026. GPT-5.2-Codex's coding index rests on 3 independent benchmarks and Qwen3.7 Max's on 3. Where only one model has a result, the page says so instead of naming a winner.