GPT-5.3-Codex vs Qwen3.5-27B
GPT-5.3-Codex leads Qwen3.5-27B on the Artificial Analysis Intelligence Index, 32.5 to 22.9, and Qwen3.5-27B costs 9.0x less per output token ($1.56 vs $14.00 per million). Updated October 22, 2026; 13 benchmarks have a result for both.
GPT-5.3-Codex or Qwen3.5-27B: which to pick
Only GPT-5.3-Codex has a coding score, 60.0 on the Vibe Coding Index; Qwen3.5-27B has none, so the two cannot be ranked on coding.
Pick Qwen3.5-27B for output tokens 9.0x cheaper ($1.56 vs $14.00 per million) and work measured by τ²-Bench Telecom or IFBench, where it leads.
- Shared results
- 13
- GPT-5.3-Codex only
- 21
- Qwen3.5-27B 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
No call
Only GPT-5.3-Codex has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
GPT-5.3-Codex
GPT-5.3-Codex leads, 32.5 against 22.9.
Same board, both measured
Web design
Design Arena: websites
No call
Only GPT-5.3-Codex has a Design Arena: websites result, 1162.
Directional only
Long documents
Listed context window
GPT-5.3-Codex
GPT-5.3-Codex takes 400K tokens against 262K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Qwen3.5-27B
GPT-5.3-Codex $0.0088, Qwen3.5-27B $0.0010 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only GPT-5.3-Codex has the number, $0.59.
Directional only
At a glance
- Coding index
- GPT-5.3-Codex 60.0 from 3 benchmarks, Qwen3.5-27B not benchmarked.
- Widest category
- Agentic work, 21.9 points apart on average over 2 shared benchmarks.
- Widest single gap
- ALE-Bench: GPT-5.3-Codex 1655.2 against Qwen3.5-27B 349.4, 43.5 points apart on a 0 to 100 scale.
- Output price
- GPT-5.3-Codex $14.00 against Qwen3.5-27B $1.56 per million output tokens, 9.0x apart.
- Context window
- GPT-5.3-Codex takes 400K tokens, Qwen3.5-27B 262K.
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-5.3-Codex | Bars | Qwen3.5-27B | Ties |
|---|---|---|---|---|
| Coding | 6 | 0 | ||
| Intelligence | 1 | 0 | ||
| Agentic work | 1 | 1 | ||
| Reasoning | 2 | 0 | ||
| General | 1 | 1 | ||
| All shared boards | 11 | 2 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index60
| Benchmark | GPT-5.3-Codex | Qwen3.5-27B | Leader |
|---|---|---|---|
| LMArena WebDev | 1407 | 1358 | GPT-5.3-Codex |
| LiveCodeBench (Vals AI) | 87.3% | Not measured | Only GPT-5.3-Codex measured |
| SWE-bench Verified (Vals AI) | 78.0% | Not measured | Only GPT-5.3-Codex measured |
| Vibe Code Bench (Vals AI) | 61.8% | Not measured | Only GPT-5.3-Codex measured |
| IOI (Vals AI) | 53.8% | Not measured | Only GPT-5.3-Codex measured |
| Terminal-Bench Hard (Artificial Analysis run) | 53.0% | 32.6% | GPT-5.3-Codex |
| Terminal-Bench | 78.4% | Not measured | Only GPT-5.3-Codex measured |
| WeirdML | 79.3% | 39.5% | GPT-5.3-Codex |
| ALE-Bench | 1655.2 | 349.4 | GPT-5.3-Codex |
| LMArena Text: coding | Not measured | 1427 | Only Qwen3.5-27B measured |
| LMArena Text: software and IT | Not measured | 1431 | Only Qwen3.5-27B measured |
| LMArena WebDev: React | 1395 | 1346 | GPT-5.3-Codex |
| LMArena WebDev: HTML | 1422 | 1386 | GPT-5.3-Codex |
| LMArena WebDev: from an image | 1443 | Not measured | Only GPT-5.3-Codex measured |
| Neither model has a result on the other 25 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | GPT-5.3-Codex | Qwen3.5-27B | Leader |
|---|---|---|---|
| AA Intelligence Index | 32.5 | 22.9 | GPT-5.3-Codex |
| Epoch Capabilities Index | 157.1 | Not measured | Only GPT-5.3-Codex measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index11
| Benchmark | GPT-5.3-Codex | Qwen3.5-27B | Leader |
|---|---|---|---|
| Design Arena: agents building web apps | 1032 | Not measured | Only GPT-5.3-Codex measured |
| Design Arena: agents building full-stack apps | 983 | Not measured | Only GPT-5.3-Codex measured |
| Design Arena: agents building mobile apps | 1053 | Not measured | Only GPT-5.3-Codex measured |
| Design Arena: agents building Android apps | 1084 | Not measured | Only GPT-5.3-Codex measured |
| Design Arena: agents building Godot games | 1124 | Not measured | Only GPT-5.3-Codex measured |
| τ²-Bench Telecom | 86.0% | 93.9% | Qwen3.5-27B |
| METR time horizons | 74.5% | Not measured | Only GPT-5.3-Codex measured |
| Vending-Bench 2 | $5,940 | $202 | GPT-5.3-Codex |
| Neither model has a result on the other 33 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | GPT-5.3-Codex | Qwen3.5-27B | Leader |
|---|---|---|---|
| GPQA Diamond (Artificial Analysis run) | 91.5% | 85.8% | GPT-5.3-Codex |
| Humanity's Last Exam (Artificial Analysis run) | 42.5% | 23.9% | GPT-5.3-Codex |
| LMArena Text: hard prompts | Not measured | 1415 | Only Qwen3.5-27B measured |
| LMArena Text: expert | Not measured | 1427 | Only Qwen3.5-27B measured |
| LMArena Text: hard prompts in English | Not measured | 1429 | Only Qwen3.5-27B measured |
| LMArena Vision: diagrams | Not measured | 1249 | Only Qwen3.5-27B measured |
| LMArena Vision: homework | Not measured | 1256 | Only Qwen3.5-27B measured |
| Neither model has a result on the other 15 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | GPT-5.3-Codex | Qwen3.5-27B | Leader |
|---|---|---|---|
| LMArena Text: math | Not measured | 1429 | Only Qwen3.5-27B measured |
| LMArena Text: mathematical work | Not measured | 1429 | Only Qwen3.5-27B measured |
| Neither model has a result on the other 11 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index4 boards, none shared
| Benchmark | GPT-5.3-Codex | Qwen3.5-27B | Leader |
|---|---|---|---|
| LMArena Text: science | Not measured | 1437 | Only Qwen3.5-27B measured |
| LMArena Text: medicine and healthcare | Not measured | 1425 | Only Qwen3.5-27B measured |
| LMArena Text: legal and government | Not measured | 1405 | Only Qwen3.5-27B measured |
| LMArena Text: business and finance | Not measured | 1399 | Only Qwen3.5-27B measured |
| Neither model has a result on the other 14 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
| Benchmark | GPT-5.3-Codex | Qwen3.5-27B | Leader |
|---|---|---|---|
| Design Arena: websites | 1162 | Not measured | Only GPT-5.3-Codex measured |
| Design Arena: UI components | 1134 | Not measured | Only GPT-5.3-Codex measured |
| Design Arena: game dev | 1161 | Not measured | Only GPT-5.3-Codex measured |
| Design Arena: data viz | 1164 | Not measured | Only GPT-5.3-Codex measured |
| Design Arena: 3D | 1007 | Not measured | Only GPT-5.3-Codex measured |
| Design Arena: all code categories | 1144 | Not measured | Only GPT-5.3-Codex measured |
| Design Arena: SVG | 1129 | Not measured | Only GPT-5.3-Codex measured |
| Design Arena: ASCII art | 1152 | Not measured | Only GPT-5.3-Codex 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-5.3-Codex | Qwen3.5-27B | Leader |
|---|---|---|---|
| IFBench | 75.4% | 75.6% | Qwen3.5-27B |
| Long context reasoning (AA-LCR) | 83.3% | 77.7% | GPT-5.3-Codex |
| DTBench | Not measured | 82.4% | Only Qwen3.5-27B measured |
| LMCA | Not measured | 34.0% | Only Qwen3.5-27B measured |
| LMArena Text | Not measured | 1409 | Only Qwen3.5-27B measured |
| LMArena Text: instruction following | Not measured | 1393 | Only Qwen3.5-27B measured |
| LMArena Text: creative writing | Not measured | 1362 | Only Qwen3.5-27B measured |
| LMArena Vision | Not measured | 1241 | Only Qwen3.5-27B measured |
| LMArena Text: English | Not measured | 1425 | Only Qwen3.5-27B measured |
| LMArena Text: non-English | Not measured | 1390 | Only Qwen3.5-27B measured |
| LMArena Text: Chinese | Not measured | 1478 | Only Qwen3.5-27B measured |
| LMArena Text: German | Not measured | 1392 | Only Qwen3.5-27B measured |
| LMArena Text: Japanese | Not measured | 1346 | Only Qwen3.5-27B measured |
| LMArena Text: Korean | Not measured | 1359 | Only Qwen3.5-27B measured |
| LMArena Text: French | Not measured | 1410 | Only Qwen3.5-27B measured |
| LMArena Text: Spanish | Not measured | 1407 | Only Qwen3.5-27B measured |
| LMArena Text: Russian | Not measured | 1390 | Only Qwen3.5-27B measured |
| LMArena Text: Polish | Not measured | 1375 | Only Qwen3.5-27B measured |
| LMArena Text: multi-turn | Not measured | 1410 | Only Qwen3.5-27B measured |
| LMArena Text: longer queries | Not measured | 1413 | Only Qwen3.5-27B measured |
| LMArena Text: writing and language | Not measured | 1380 | Only Qwen3.5-27B measured |
| LMArena Text: entertainment and media | Not measured | 1363 | Only Qwen3.5-27B measured |
| LMArena Vision: English | Not measured | 1246 | Only Qwen3.5-27B measured |
| LMArena Vision: Chinese | Not measured | 1301 | Only Qwen3.5-27B measured |
| LMArena Vision: reading text | Not measured | 1246 | Only Qwen3.5-27B measured |
| LMArena Vision: humor | Not measured | 1227 | Only Qwen3.5-27B measured |
| LMArena Vision: captioning | Not measured | 1170 | Only Qwen3.5-27B measured |
| LMArena Vision: recognizing things | Not measured | 1210 | Only Qwen3.5-27B measured |
| LMArena Vision: creative writing | Not measured | 1224 | Only Qwen3.5-27B 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 | GPT-5.3-Codex | Qwen3.5-27B | 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
Qwen3.5-27B costs less
Repository review
50K input + 3K output tokens
Qwen3.5-27B costs less
Agent loop
200K input + 10K output tokens
Qwen3.5-27B 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-5.3-Codex | Qwen3.5-27B |
|---|---|---|
| Maker | OpenAI | Qwen |
| Weights | No public weights listed | Open weights |
| Released | February 24, 2026 | February 25, 2026 |
| Context window | 400K | 262K |
| Input $/Mtok | $1.75 | $0.195 |
| Output $/Mtok | $14.00 | $1.56 |
| Coding index | 60.0 | Not benchmarked |
| Independent measurements behind it | 3 | 1 |
| Cost per solved task | $0.59 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GPT-5.3-Codex better than Qwen3.5-27B for coding?
Only GPT-5.3-Codex has a coding score, 60.0 on the Vibe Coding Index; Qwen3.5-27B has none, so the two cannot be ranked on coding. Pick Qwen3.5-27B for output tokens 9.0x cheaper ($1.56 vs $14.00 per million) and work measured by τ²-Bench Telecom or IFBench, where it leads.
Which is cheaper, GPT-5.3-Codex or Qwen3.5-27B?
GPT-5.3-Codex costs $1.75 per million input tokens and $14.00 per million output tokens; Qwen3.5-27B costs $0.195 and $1.56. Cost per solved task is known for GPT-5.3-Codex only.
Which has the bigger context window, GPT-5.3-Codex or Qwen3.5-27B?
GPT-5.3-Codex, with 400K tokens against 262K.
How many benchmarks is the GPT-5.3-Codex vs Qwen3.5-27B comparison based on?
13 benchmarks have a result for both models as of October 22, 2026. GPT-5.3-Codex's coding index rests on 3 independent benchmarks and Qwen3.5-27B's on 1. Where only one model has a result, the page says so instead of naming a winner.