GPT-5.2 vs GPT-5.4
GPT-5.4 leads GPT-5.2 on coding, 64.4 to 55.9 on the Vibe Coding Index, at about the same output price ($15.00 vs $14.00 per million). Updated October 22, 2026; 100 benchmarks have a result for both.
GPT-5.2 or GPT-5.4: which to pick
Pick GPT-5.4 for coding: it scores 64.4 on the Vibe Coding Index against 55.9 for GPT-5.2.
Pick GPT-5.2 for output tokens 1.1x cheaper ($14.00 vs $15.00 per million) and work measured by LiveCodeBench (Vals AI) or Design Arena: agents building web apps, where it leads.
- Shared results
- 100
- GPT-5.2 only
- 14
- GPT-5.4 only
- 10
- Like-for-like categories
- 8 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-5.4
GPT-5.4 scores 64.4 against 55.9.
3 benchmarks behind each score
Intelligence
AA Intelligence Index
GPT-5.4
GPT-5.4 leads, 39.0 against 30.4.
Same board, both measured
Reasoning
GPQA Diamond
No call
Tied at 91.7%.
Same board, both measured
Math
AIME
GPT-5.2
GPT-5.2 leads, 96.9% against 96.7%.
Same board, both measured
Knowledge
MMLU Pro
GPT-5.4
GPT-5.4 leads, 87.5% against 86.2%.
Same board, both measured
Web design
Design Arena: websites
GPT-5.4
GPT-5.4 leads, 1198 against 1195.
Same board, both measured
Long documents
Listed context window
GPT-5.4
GPT-5.4 takes 1.05M tokens against 400K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
GPT-5.2
GPT-5.2 $0.0088, GPT-5.4 $0.010 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
GPT-5.4
GPT-5.2 $1.91, GPT-5.4 $1.02.
Published attempt cost and pass rate
At a glance
- Coding index
- GPT-5.2 55.9 from 3 benchmarks, GPT-5.4 64.4 from 3 benchmarks.
- Widest category
- Math, 12.6 points apart on average over 7 shared benchmarks.
- Widest single gap
- ProofBench: GPT-5.2 15.0% against GPT-5.4 56.0%, 41.0 points apart on a 0 to 100 scale.
- Output price
- GPT-5.2 $14.00 against GPT-5.4 $15.00 per million output tokens, 1.1x apart.
- Context window
- GPT-5.2 takes 400K tokens, GPT-5.4 1.05M.
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 | Bars | GPT-5.4 | Ties |
|---|---|---|---|---|
| Coding | 1 | 11 | ||
| Intelligence | 0 | 2 | ||
| Agentic work | 7 | 2 | ||
| Reasoning | 1 | 13 | 2 ties | |
| Math | 1 | 6 | ||
| Knowledge | 4 | 9 | ||
| Web design | 3 | 5 | ||
| General | 5 | 28 | ||
| All shared boards | 22 | 76 | 2 ties |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index111
| Benchmark | GPT-5.2 | GPT-5.4 | Leader |
|---|---|---|---|
| SWE-bench Verified | 72.8% | Not measured | Only GPT-5.2 measured |
| SWE-bench Multilingual | 66.7% | Not measured | Only GPT-5.2 measured |
| LMArena WebDev | 1417 | 1464 | GPT-5.4 |
| AA Coding Index | Not measured | 71.1 | Only GPT-5.4 measured |
| LiveCodeBench (Vals AI) | 85.4% | 84.1% | GPT-5.2 |
| SWE-bench Verified (Vals AI) | 75.8% | 78.2% | GPT-5.4 |
| Vibe Code Bench (Vals AI) | 53.5% | 67.4% | GPT-5.4 |
| Code migration (Vals AI) | Not measured | 35.0% | Only GPT-5.4 measured |
| ProgramBench (Vals AI) | Not measured | 0.5% | Only GPT-5.4 measured |
| SkillsBench (Vals AI) | Not measured | 51.7% | Only GPT-5.4 measured |
| LiveCodeBench (Artificial Analysis run) | 89.4% | Not measured | Only GPT-5.2 measured |
| Terminal-Bench Hard (Artificial Analysis run) | 47.0% | 57.6% | GPT-5.4 |
| Terminal-Bench | 64.9% | 81.8% | GPT-5.4 |
| GSO-Bench | 27.5% | 31.4% | GPT-5.4 |
| WeirdML | 72.2% | 77.7% | GPT-5.4 |
| MirrorCode | Not measured | 15.6% | Only GPT-5.4 measured |
| ALE-Bench | 1293.5 | 1607.0 | GPT-5.4 |
| LMArena Text: coding | 1442 | 1496 | GPT-5.4 |
| LMArena Text: software and IT | 1434 | 1491 | GPT-5.4 |
| LMArena WebDev: React | Not measured | 1454 | Only GPT-5.4 measured |
| LMArena WebDev: HTML | 1427 | 1478 | GPT-5.4 |
| LMArena WebDev: from an image | Not measured | 1436 | Only GPT-5.4 measured |
| Neither model has a result on the other 17 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | GPT-5.2 | GPT-5.4 | Leader |
|---|---|---|---|
| AA Intelligence Index | 30.4 | 39.0 | GPT-5.4 |
| Epoch Capabilities Index | 153.7 | 157.3 | GPT-5.4 |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index72
| Benchmark | GPT-5.2 | GPT-5.4 | Leader |
|---|---|---|---|
| Design Arena: agents building web apps | 1086 | 1059 | GPT-5.2 |
| Design Arena: agents building full-stack apps | 1043 | 1017 | GPT-5.2 |
| Design Arena: agents building mobile apps | 1093 | 1087 | GPT-5.2 |
| Design Arena: agents building Android apps | 1059 | 1031 | GPT-5.2 |
| Design Arena: agents building Godot games | 1142 | 1135 | GPT-5.2 |
| τ²-Bench Telecom | 84.8% | 87.1% | GPT-5.4 |
| APEX-Agents | Not measured | 52.4% | Only GPT-5.4 measured |
| METR time horizons | 75.3% | 74.3% | GPT-5.2 |
| Remote Labor Index | 2.5% | Not measured | Only GPT-5.2 measured |
| GDPval | 49.7% | Not measured | Only GPT-5.2 measured |
| DeepResearch Bench | 41.1% | 35.1% | GPT-5.2 |
| PostTrainBench | Not measured | 19.0% | Only GPT-5.4 measured |
| Vending-Bench 2 | $3,591 | $6,144 | GPT-5.4 |
| Berkeley Function Calling Leaderboard (BFCL) | 55.9% | Not measured | Only GPT-5.2 measured |
| Neither model has a result on the other 27 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index113
| Benchmark | GPT-5.2 | GPT-5.4 | Leader |
|---|---|---|---|
| GPQA Diamond | 91.7% | 91.7% | Tie |
| MMMU Pro (Vals AI) | 86.7% | 87.5% | GPT-5.4 |
| GPQA Diamond (Artificial Analysis run) | 90.3% | 92.0% | GPT-5.4 |
| Humanity's Last Exam (Artificial Analysis run) | 37.7% | 43.7% | GPT-5.4 |
| GPQA Diamond (Epoch AI run) | 91.4% | 93.3% | GPT-5.4 |
| ARC-AGI | 86.2% | 93.7% | GPT-5.4 |
| ARC-AGI-2 | 52.9% | 74.0% | GPT-5.4 |
| SimpleBench | 45.8% | Not measured | Only GPT-5.2 measured |
| Chess puzzles | 48.0% | 48.0% | Tie |
| Mystery game puzzles | 23.0% | 37.0% | GPT-5.4 |
| EBR-bench | 23.0% | 25.4% | GPT-5.4 |
| VPCT | 84.0% | Not measured | Only GPT-5.2 measured |
| LMArena Text: hard prompts | 1428 | 1485 | GPT-5.4 |
| LMArena Text: expert | 1445 | 1507 | GPT-5.4 |
| LMArena Text: hard prompts in English | 1431 | 1479 | GPT-5.4 |
| LMArena Vision: diagrams | 1265 | 1327 | GPT-5.4 |
| LMArena Vision: homework | 1276 | 1336 | GPT-5.4 |
| Kagi LLM Benchmark | 73.3% | 63.8% | GPT-5.2 |
| Neither model has a result on the other 4 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index16
| Benchmark | GPT-5.2 | GPT-5.4 | Leader |
|---|---|---|---|
| AIME | 96.9% | 96.7% | GPT-5.2 |
| MGSM (Vals AI) | 94.0% | Not measured | Only GPT-5.2 measured |
| AA Math Index | 99.0 | Not measured | Only GPT-5.2 measured |
| AIME 2025 (Artificial Analysis run) | 99.0% | Not measured | Only GPT-5.2 measured |
| OTIS Mock AIME | 96.1% | 97.8% | GPT-5.4 |
| FrontierMath, tiers 1 to 3 | 67.4% | 78.6% | GPT-5.4 |
| FrontierMath, tier 4 | 31.7% | 49.0% | GPT-5.4 |
| ProofBench | 15.0% | 56.0% | GPT-5.4 |
| LMArena Text: math | 1440 | 1488 | GPT-5.4 |
| LMArena Text: mathematical work | 1442 | 1497 | GPT-5.4 |
| Neither model has a result on the other 3 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index49
| Benchmark | GPT-5.2 | GPT-5.4 | Leader |
|---|---|---|---|
| MMLU Pro | 86.2% | 87.5% | GPT-5.4 |
| LegalBench (Vals AI) | 82.8% | 86.0% | GPT-5.4 |
| CaseLaw (Vals AI) | 66.0% | 63.8% | GPT-5.2 |
| CorpFin (Vals AI) | 65.9% | 65.3% | GPT-5.2 |
| TaxEval (Vals AI) | 75.8% | 74.0% | GPT-5.2 |
| MortgageTax (Vals AI) | 67.1% | 68.3% | GPT-5.4 |
| MedQA (Vals AI) | 94.1% | 96.1% | GPT-5.4 |
| MedCode (Vals AI) | 49.8% | 41.3% | GPT-5.2 |
| MMLU Pro (Artificial Analysis run) | 87.4% | Not measured | Only GPT-5.2 measured |
| SimpleQA Verified | 37.1% | 45.1% | GPT-5.4 |
| LMArena Text: science | 1425 | 1479 | GPT-5.4 |
| LMArena Text: medicine and healthcare | 1425 | 1458 | GPT-5.4 |
| LMArena Text: legal and government | 1427 | 1486 | GPT-5.4 |
| LMArena Text: business and finance | 1414 | 1474 | GPT-5.4 |
| FACTS Benchmark Suite | 48.5% | Not measured | Only GPT-5.2 measured |
| Neither model has a result on the other 3 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index35
| Benchmark | GPT-5.2 | GPT-5.4 | Leader |
|---|---|---|---|
| Design Arena: websites | 1195 | 1198 | GPT-5.4 |
| Design Arena: UI components | 1190 | 1212 | GPT-5.4 |
| Design Arena: game dev | 1195 | 1194 | GPT-5.2 |
| Design Arena: data viz | 1202 | 1196 | GPT-5.2 |
| Design Arena: 3D | 1079 | 1065 | GPT-5.2 |
| Design Arena: all code categories | 1170 | 1179 | GPT-5.4 |
| Design Arena: SVG | 1139 | 1141 | GPT-5.4 |
| Design Arena: ASCII art | 1152 | 1188 | GPT-5.4 |
| Neither model has a result on the other 1 board here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index528
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-5.2 | GPT-5.4 | 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
GPT-5.2 costs less
Repository review
50K input + 3K output tokens
GPT-5.2 costs less
Agent loop
200K input + 10K output tokens
GPT-5.2 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 | OpenAI GPT-5.4 |
|---|---|---|
| Maker | OpenAI | OpenAI |
| Weights | No public weights listed | No public weights listed |
| Released | December 10, 2025 | March 5, 2026 |
| Context window | 400K | 1.05M |
| Input $/Mtok | $1.75 | $2.50 |
| Output $/Mtok | $14.00 | $15.00 |
| Coding index | 55.9 | 64.4 |
| Independent measurements behind it | 3 | 3 |
| Cost per solved task | $1.91 | $1.02 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GPT-5.2 better than GPT-5.4 for coding?
Pick GPT-5.4 for coding: it scores 64.4 on the Vibe Coding Index against 55.9 for GPT-5.2. Pick GPT-5.2 for output tokens 1.1x cheaper ($14.00 vs $15.00 per million) and work measured by LiveCodeBench (Vals AI) or Design Arena: agents building web apps, where it leads.
Which is cheaper, GPT-5.2 or GPT-5.4?
GPT-5.2 costs $1.75 per million input tokens and $14.00 per million output tokens; GPT-5.4 costs $2.50 and $15.00. Per solved SWE-bench Verified task, GPT-5.2 costs $1.91 and GPT-5.4 $1.02.
Which has the bigger context window, GPT-5.2 or GPT-5.4?
GPT-5.4, with 1.05M tokens against 400K.
How many benchmarks is the GPT-5.2 vs GPT-5.4 comparison based on?
100 benchmarks have a result for both models as of October 22, 2026. GPT-5.2's coding index rests on 3 independent benchmarks and GPT-5.4's on 3. Where only one model has a result, the page says so instead of naming a winner.