Model comparison
GPT-4 vs GPT-5.2
GPT-5.2 is the stronger model overall, scoring 54.1 to 29.1 on the Noometry Index.
Last verified . 28 shared benchmarks.
Summary
- They share 28 benchmarks with published results for both. GPT-4 scores higher in 0 categories and GPT-5.2 in 8 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in math, where GPT-5.2 leads 60.0 to 10.8.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 1.1% for GPT-4 and 96.1% for GPT-5.2.
- GPT-5.2 is cheaper at $1.75 / $14 per million input/output tokens, against $30 / $60 for GPT-4.
- GPT-5.2 accepts more context: 400K tokens versus 8K.
Side by side
| GPT-4 | GPT-5.2 | |
|---|---|---|
| Provider | OpenAI | OpenAI |
| Noometry Index | 29.1 | 54.1 |
| Released | 2023-03-14 | 2025-12-11 |
| Weights | Proprietary | Proprietary |
| Context window | 8K | 400K |
| Max output | 8K | 128K |
| Input $ / M tokens | $30 | $1.75 |
| Output $ / M tokens | $60 | $14 |
| Results tracked | 38 | 67 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5.2 leads
GPT-4: 31.6 (#283), GPT-5.2: 51.6 (#37)
| Benchmark | GPT-4 | GPT-5.2 |
|---|---|---|
| WeirdML | 12.4% | 72.2% |
| LMArena Coding | 1254 | 1447 |
| SWE-bench Verified | — | 73.8% |
| SWE-bench Verified (bash only) | — | 72.8% |
| LMArena WebDev | — | 1416 |
| SWE-bench Multilingual | — | 66.7% |
| GSO | — | 27.4% |
| BigCodeBench Instruct | 46% | — |
| BigCodeBench Complete | 57.2% | — |
| ALE-Bench | — | 1,294 |
| AlgoTune | — | 2.05 |
| HumanEval+ | 79.3% | — |
Agentic & Tool Use Not comparable
GPT-4: —, GPT-5.2: 40.2 (#24)
| Benchmark | GPT-4 | GPT-5.2 |
|---|---|---|
| METR Time Horizons | 36.1% | 75.3% |
| Terminal-Bench | — | 64.9% |
| Berkeley Function Calling Leaderboard | — | 55.9% |
| GDPval | — | 49.7% |
| Remote Labor Index | — | 2.5% |
| τ²-bench Airline | — | 83% |
| τ²-bench Banking | — | 32.2% |
| τ²-bench Retail | — | 81.6% |
| τ²-bench Telecom | — | 89.7% |
| DeepResearch Bench | — | 41.1% |
| LMArena Search | — | 1207 |
| Vending-Bench 2 | — | 3,591 |
Reasoning GPT-5.2 leads
GPT-4: 17.8 (#289), GPT-5.2: 50.2 (#35)
| Benchmark | GPT-4 | GPT-5.2 |
|---|---|---|
| Chess Puzzles | 4% | 49% |
| LMArena Hard Prompts | 1241 | 1445 |
| Mystery Game Puzzles | 12% | 23% |
| DTBench | 62.7% | 90.9% |
| LMCA | 17.1% | 43.9% |
| Epoch Capabilities Index | 125.89 | 153.45 |
| ForecastBench | 57.8 | 60.1 |
| ARC-AGI-2 | — | 52.9% |
| SimpleBench | — | 45.8% |
| Kagi LLM Benchmark | — | 73.3% |
| NYT Connections (extended) | — | 83.6% |
| ARC-AGI-1 | — | 86.2% |
| EnigmaEval | — | 10.4% |
| EBR-Bench | — | 23% |
| BIG-Bench Hard | 75.1% | — |
| HellaSwag | 95.3% | — |
| WinoGrande | 87.5% | — |
Math GPT-5.2 leads
GPT-4: 10.8 (#309), GPT-5.2: 60.0 (#38)
| Benchmark | GPT-4 | GPT-5.2 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 1.1% | 96.1% |
| LMArena Math | 1269 | 1440 |
| FrontierMath (Tiers 1-3) | — | 67.4% |
| FrontierMath Tier 4 | — | 31.7% |
| MathArena Final-Answer Competitions | — | 72% |
| ProofBench | — | 15% |
| MATH Level 5 | 23% | — |
| FrontierMath (Feb 2025 set) | — | 40.7% |
| FrontierMath Tier 4 (v1) | — | 18.8% |
| GSM8K | 92% | — |
Knowledge GPT-5.2 leads
GPT-4: 18.4 (#282), GPT-5.2: 59.3 (#32)
| Benchmark | GPT-4 | GPT-5.2 |
|---|---|---|
| GPQA Diamond | 35.7% | 91.4% |
| LMArena Expert | 1211 | 1445 |
| Humanity's Last Exam | — | 27.8% |
| SimpleQA Verified | — | 37.1% |
| Vectara Hallucination Rate | — | 8.4% |
| MMLU | 86.4% | — |
| TriviaQA | 84.8% | — |
Multimodal Not comparable
GPT-4: —, GPT-5.2: 51.3 (#7)
| Benchmark | GPT-4 | GPT-5.2 |
|---|---|---|
| LMArena Vision | — | 1268 |
| VPCT | — | 84% |
| Furniture Assembly | — | 38.3% |
| LMArena Document | — | 1405 |
Multilingual GPT-5.2 leads
GPT-4: 40.6 (#215), GPT-5.2: 53.4 (#67)
| Benchmark | GPT-4 | GPT-5.2 |
|---|---|---|
| LMArena Non-English | 1246 | 1425 |
| LMArena Chinese | 1242 | 1460 |
| LMArena French | 1283 | 1455 |
| LMArena German | 1251 | 1448 |
| LMArena Japanese | 1209 | 1420 |
| LMArena Korean | 1184 | 1392 |
| LMArena Russian | 1251 | 1440 |
| LMArena Spanish | 1261 | 1433 |
Instruction Following GPT-5.2 leads
GPT-4: 65.3 (#222), GPT-5.2: 74.7 (#89)
| Benchmark | GPT-4 | GPT-5.2 |
|---|---|---|
| LMArena Instruction Following | 1241 | 1417 |
Long Context GPT-5.2 leads
GPT-4: 37.7 (#212), GPT-5.2: 44.0 (#78)
| Benchmark | GPT-4 | GPT-5.2 |
|---|---|---|
| LMArena Longer Query | 1244 | 1428 |
| CL-bench | — | 18.2% |
Writing & Preference GPT-5.2 leads
GPT-4: 34.9 (#268), GPT-5.2: 66.8 (#32)
| Benchmark | GPT-4 | GPT-5.2 |
|---|---|---|
| LMArena Text | 1263 | 1439 |
| LMArena Creative Writing | 1244 | 1401 |
| EQ-Bench Creative Writing | 752 | 1703 |
| LMArena Multi-Turn | 1257 | 1458 |
Frequently asked questions
Is GPT-4 better than GPT-5.2?
GPT-5.2 is the stronger model overall, scoring 54.1 to 29.1 on the Noometry Index.
Which is cheaper, GPT-4 or GPT-5.2?
GPT-5.2 is cheaper. It lists at $1.75 per million input tokens and $14 per million output tokens; GPT-4 lists at $30 and $60.
Is GPT-4 or GPT-5.2 better for coding?
GPT-5.2 scores higher on coding benchmarks: 51.6 versus 31.6 in the Noometry coding category.
Which has the bigger context window?
GPT-5.2 does, with 400K tokens against 8K.
How many benchmarks do GPT-4 and GPT-5.2 share?
28 benchmarks have published results for both models. GPT-4 has 38 scored results on Noometry and GPT-5.2 has 67.