GPT-5.2-Codex vs GPT-5.3-Codex
GPT-5.3-Codex leads GPT-5.2-Codex on coding, 60.0 to 45.4 on the Vibe Coding Index, at about the same output price ($14.00 vs $14.00 per million). Updated October 21, 2026; 20 benchmarks have a result for both.
GPT-5.2-Codex or GPT-5.3-Codex: which to pick
Pick GPT-5.3-Codex for coding: it scores 60.0 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 $0.59) and work measured by LiveCodeBench (Vals AI) or Design Arena: agents building web apps, where it leads.
- Shared results
- 20
- GPT-5.2-Codex only
- 2
- GPT-5.3-Codex only
- 14
- 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
GPT-5.3-Codex
GPT-5.3-Codex scores 60.0 against 45.4.
3 benchmarks behind each score
Intelligence
AA Intelligence Index
GPT-5.3-Codex
GPT-5.3-Codex leads, 32.5 against 28.5.
Same board, both measured
Web design
Design Arena: websites
No call
Only GPT-5.3-Codex has a Design Arena: websites result, 1163.
Directional only
Long documents
Listed context window
No call
Both take 400K tokens.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
No call
GPT-5.2-Codex $0.0088, GPT-5.3-Codex $0.0088 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
GPT-5.2-Codex
GPT-5.2-Codex $0.43, GPT-5.3-Codex $0.59.
Published attempt cost and pass rate
At a glance
- Coding index
- GPT-5.2-Codex 45.4 from 3 benchmarks, GPT-5.3-Codex 60.0 from 3 benchmarks.
- Widest category
- Coding, 11.4 points apart on average over 9 shared benchmarks.
- Widest single gap
- Vibe Code Bench (Vals AI): GPT-5.2-Codex 37.9% against GPT-5.3-Codex 61.8%, 23.9 points apart on a 0 to 100 scale.
- Output price
- Both cost $14.00 per million output tokens.
- Context window
- Both take 400K 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-5.2-Codex | Bars | GPT-5.3-Codex | Ties |
|---|---|---|---|---|
| Coding | 1 | 8 | ||
| Intelligence | 0 | 1 | ||
| Agentic work | 6 | 0 | ||
| Reasoning | 0 | 2 | ||
| General | 1 | 1 | ||
| All shared boards | 8 | 12 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index18
| Benchmark | GPT-5.2-Codex | GPT-5.3-Codex | 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 | 1407 | GPT-5.3-Codex |
| LiveCodeBench (Vals AI) | 88.0% | 87.3% | GPT-5.2-Codex |
| SWE-bench Verified (Vals AI) | 72.4% | 78.0% | GPT-5.3-Codex |
| Vibe Code Bench (Vals AI) | 37.9% | 61.8% | GPT-5.3-Codex |
| IOI (Vals AI) | Not measured | 53.8% | Only GPT-5.3-Codex measured |
| Terminal-Bench Hard (Artificial Analysis run) | 37.1% | 53.0% | GPT-5.3-Codex |
| Terminal-Bench | 66.5% | 78.4% | GPT-5.3-Codex |
| WeirdML | Not measured | 79.3% | Only GPT-5.3-Codex measured |
| ALE-Bench | 1299.9 | 1655.2 | GPT-5.3-Codex |
| LMArena WebDev: React | 1329 | 1395 | GPT-5.3-Codex |
| LMArena WebDev: HTML | 1348 | 1422 | GPT-5.3-Codex |
| LMArena WebDev: from an image | Not measured | 1443 | 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 index01
| Benchmark | GPT-5.2-Codex | GPT-5.3-Codex | Leader |
|---|---|---|---|
| AA Intelligence Index | 28.5 | 32.5 | GPT-5.3-Codex |
| Epoch Capabilities Index | Not measured | 157.1 | Only GPT-5.3-Codex measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index60
| Benchmark | GPT-5.2-Codex | GPT-5.3-Codex | Leader |
|---|---|---|---|
| Design Arena: agents building web apps | 1051 | 1032 | GPT-5.2-Codex |
| Design Arena: agents building full-stack apps | 993 | 983 | GPT-5.2-Codex |
| Design Arena: agents building mobile apps | 1090 | 1054 | GPT-5.2-Codex |
| Design Arena: agents building Android apps | 1176 | 1084 | GPT-5.2-Codex |
| Design Arena: agents building Godot games | 1142 | 1124 | GPT-5.2-Codex |
| τ²-Bench Telecom | 92.1% | 86.0% | GPT-5.2-Codex |
| METR time horizons | Not measured | 74.5% | Only GPT-5.3-Codex measured |
| Vending-Bench 2 | Not measured | $5,940 | Only GPT-5.3-Codex measured |
| Neither model has a result on the other 33 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | GPT-5.2-Codex | GPT-5.3-Codex | Leader |
|---|---|---|---|
| GPQA Diamond (Artificial Analysis run) | 89.9% | 91.5% | GPT-5.3-Codex |
| Humanity's Last Exam (Artificial Analysis run) | 35.7% | 42.5% | GPT-5.3-Codex |
| Neither model has a result on the other 20 boards here | |||
MathNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-5.2-Codex | GPT-5.3-Codex | Leader |
|---|---|---|---|
| Neither model has a result on 13 boards here | |||
KnowledgeNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-5.2-Codex | GPT-5.3-Codex | Leader |
|---|---|---|---|
| Neither model has a result on 18 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
| Benchmark | GPT-5.2-Codex | GPT-5.3-Codex | Leader |
|---|---|---|---|
| Design Arena: websites | Not measured | 1163 | Only GPT-5.3-Codex measured |
| Design Arena: UI components | Not measured | 1137 | Only GPT-5.3-Codex measured |
| Design Arena: game dev | Not measured | 1161 | Only GPT-5.3-Codex measured |
| Design Arena: data viz | Not measured | 1164 | Only GPT-5.3-Codex measured |
| Design Arena: 3D | Not measured | 1008 | Only GPT-5.3-Codex measured |
| Design Arena: all code categories | Not measured | 1145 | Only GPT-5.3-Codex measured |
| Design Arena: SVG | Not measured | 1129 | Only GPT-5.3-Codex measured |
| Design Arena: ASCII art | Not measured | 1152 | 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-5.2-Codex | GPT-5.3-Codex | Leader |
|---|---|---|---|
| IFBench | 77.6% | 75.4% | GPT-5.2-Codex |
| Long context reasoning (AA-LCR) | 82.3% | 83.3% | GPT-5.3-Codex |
| Neither model has a result on the other 44 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-5.2-Codex | GPT-5.3-Codex | 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
Same cost
Repository review
50K input + 3K output tokens
Same cost
Agent loop
200K input + 10K output tokens
Same cost
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 | OpenAI GPT-5.3-Codex |
|---|---|---|
| Maker | OpenAI | OpenAI |
| Weights | No public weights listed | No public weights listed |
| Released | January 14, 2026 | February 24, 2026 |
| Context window | 400K | 400K |
| Input $/Mtok | $1.75 | $1.75 |
| Output $/Mtok | $14.00 | $14.00 |
| Coding index | 45.4 | 60.0 |
| Independent measurements behind it | 3 | 3 |
| Cost per solved task | $0.43 | $0.59 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GPT-5.2-Codex better than GPT-5.3-Codex for coding?
Pick GPT-5.3-Codex for coding: it scores 60.0 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 $0.59) and work measured by LiveCodeBench (Vals AI) or Design Arena: agents building web apps, where it leads.
Which is cheaper, GPT-5.2-Codex or GPT-5.3-Codex?
GPT-5.2-Codex costs $1.75 per million input tokens and $14.00 per million output tokens; GPT-5.3-Codex costs $1.75 and $14.00. Per solved SWE-bench Verified task, GPT-5.2-Codex costs $0.43 and GPT-5.3-Codex $0.59.
Which has the bigger context window, GPT-5.2-Codex or GPT-5.3-Codex?
Neither: both take 400K tokens.
How many benchmarks is the GPT-5.2-Codex vs GPT-5.3-Codex comparison based on?
20 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 GPT-5.3-Codex's on 3. Where only one model has a result, the page says so instead of naming a winner.