Model comparison
GPT-5.2 vs GPT-5.4 mini
GPT-5.2 is the stronger model overall, scoring 54.1 to 45.0 on the Noometry Index. GPT-5.4 mini costs 2.9× less per token, which makes it the better buy when GPT-5.2's lead doesn't matter for your workload.
Last verified . 42 shared benchmarks.
Summary
- They share 42 benchmarks with published results for both. GPT-5.2 scores higher in 10 categories and GPT-5.4 mini in 0 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where GPT-5.2 leads 50.2 to 30.4.
- The biggest single-benchmark swing is Kagi LLM Benchmark: 73.3% for GPT-5.2 and 37.9% for GPT-5.4 mini.
- GPT-5.4 mini is cheaper at $0.75 / $4.50 per million input/output tokens, against $1.75 / $14 for GPT-5.2.
Side by side
| GPT-5.2 | GPT-5.4 mini | |
|---|---|---|
| Provider | OpenAI | OpenAI |
| Noometry Index | 54.1 | 45.0 |
| Released | 2025-12-11 | 2026-03-17 |
| Weights | Proprietary | Proprietary |
| Context window | 400K | 400K |
| Max output | 128K | 128K |
| Input $ / M tokens | $1.75 | $0.75 |
| Output $ / M tokens | $14 | $4.50 |
| Results tracked | 67 | 46 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5.2 leads
GPT-5.2: 51.6 (#37), GPT-5.4 mini: 45.2 (#72)
| Benchmark | GPT-5.2 | GPT-5.4 mini |
|---|---|---|
| LMArena WebDev | 1416 | 1397 |
| WeirdML | 72.2% | 60.3% |
| LMArena Coding | 1447 | 1438 |
| ALE-Bench | 1,294 | 1,189 |
| SWE-bench Verified | 73.8% | — |
| FrontierCode | — | 27% |
| SWE-bench Verified (bash only) | 72.8% | — |
| SWE-bench Multilingual | 66.7% | — |
| SciCode | — | 49.9% |
| GSO | 27.4% | — |
| AlgoTune | 2.05 | — |
Agentic & Tool Use GPT-5.2 leads
GPT-5.2: 40.2 (#24), GPT-5.4 mini: 29.9 (#81)
| Benchmark | GPT-5.2 | GPT-5.4 mini |
|---|---|---|
| DeepResearch Bench | 41.1% | 36.3% |
| 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% | — |
| LMArena Search | 1207 | — |
| METR Time Horizons | 75.3% | — |
| Vending-Bench 2 | 3,591 | — |
Reasoning GPT-5.2 leads
GPT-5.2: 50.2 (#35), GPT-5.4 mini: 30.4 (#85)
| Benchmark | GPT-5.2 | GPT-5.4 mini |
|---|---|---|
| ARC-AGI-2 | 52.9% | 18.9% |
| Kagi LLM Benchmark | 73.3% | 37.9% |
| NYT Connections (extended) | 83.6% | 61.8% |
| ARC-AGI-1 | 86.2% | 63.7% |
| Chess Puzzles | 49% | 24% |
| LMArena Hard Prompts | 1445 | 1424 |
| Mystery Game Puzzles | 23% | 11% |
| DTBench | 90.9% | 80% |
| LMCA | 43.9% | 40.8% |
| Epoch Capabilities Index | 153.45 | 148.84 |
| ForecastBench | 60.1 | 57 |
| SimpleBench | 45.8% | — |
| CritPt | — | 10% |
| EnigmaEval | 10.4% | — |
| Thematic Generalization | — | 61.7% |
| EBR-Bench | 23% | — |
Math GPT-5.2 leads
GPT-5.2: 60.0 (#38), GPT-5.4 mini: 45.5 (#75)
| Benchmark | GPT-5.2 | GPT-5.4 mini |
|---|---|---|
| FrontierMath (Tiers 1-3) | 67.4% | 51.2% |
| FrontierMath Tier 4 | 31.7% | 9.8% |
| OTIS Mock AIME 2024-2025 | 96.1% | 88.9% |
| ProofBench | 15% | 21% |
| LMArena Math | 1440 | 1419 |
| FrontierMath (Feb 2025 set) | 40.7% | 28.3% |
| FrontierMath Tier 4 (v1) | 18.8% | 2.1% |
| MathArena Final-Answer Competitions | 72% | — |
Knowledge GPT-5.2 leads
GPT-5.2: 59.3 (#32), GPT-5.4 mini: 51.5 (#67)
| Benchmark | GPT-5.2 | GPT-5.4 mini |
|---|---|---|
| GPQA Diamond | 91.4% | 86.9% |
| SimpleQA Verified | 37.1% | 29.4% |
| Vectara Hallucination Rate | 8.4% | 5.5% |
| LMArena Expert | 1445 | 1435 |
| Humanity's Last Exam | 27.8% | — |
Multimodal GPT-5.2 leads
GPT-5.2: 51.3 (#7), GPT-5.4 mini: 39.7 (#56)
| Benchmark | GPT-5.2 | GPT-5.4 mini |
|---|---|---|
| LMArena Vision | 1268 | 1245 |
| VPCT | 84% | — |
| Furniture Assembly | 38.3% | — |
| LMArena Document | 1405 | — |
Multilingual GPT-5.2 leads
GPT-5.2: 53.4 (#67), GPT-5.4 mini: 51.9 (#96)
| Benchmark | GPT-5.2 | GPT-5.4 mini |
|---|---|---|
| LMArena Non-English | 1425 | 1405 |
| LMArena Chinese | 1460 | 1446 |
| LMArena French | 1455 | 1440 |
| LMArena German | 1448 | 1409 |
| LMArena Japanese | 1420 | 1374 |
| LMArena Korean | 1392 | 1368 |
| LMArena Russian | 1440 | 1417 |
| LMArena Spanish | 1433 | 1405 |
Instruction Following Too close to call
GPT-5.2: 74.7 (#89), GPT-5.4 mini: 74.1 (#102)
| Benchmark | GPT-5.2 | GPT-5.4 mini |
|---|---|---|
| LMArena Instruction Following | 1417 | 1405 |
Long Context GPT-5.2 leads
GPT-5.2: 44.0 (#78), GPT-5.4 mini: 43.0 (#112)
| Benchmark | GPT-5.2 | GPT-5.4 mini |
|---|---|---|
| LMArena Longer Query | 1428 | 1407 |
| CL-bench | 18.2% | — |
Writing & Preference GPT-5.2 leads
GPT-5.2: 66.8 (#32), GPT-5.4 mini: 64.0 (#58)
| Benchmark | GPT-5.2 | GPT-5.4 mini |
|---|---|---|
| LMArena Text | 1439 | 1412 |
| LMArena Creative Writing | 1401 | 1370 |
| EQ-Bench Creative Writing | 1703 | 1665 |
| LMArena Multi-Turn | 1458 | 1429 |
Frequently asked questions
Is GPT-5.2 better than GPT-5.4 mini?
GPT-5.2 is the stronger model overall, scoring 54.1 to 45.0 on the Noometry Index. GPT-5.4 mini costs 2.9× less per token, which makes it the better buy when GPT-5.2's lead doesn't matter for your workload.
Which is cheaper, GPT-5.2 or GPT-5.4 mini?
GPT-5.4 mini is cheaper. It lists at $0.75 per million input tokens and $4.50 per million output tokens; GPT-5.2 lists at $1.75 and $14.
Is GPT-5.2 or GPT-5.4 mini better for coding?
GPT-5.2 scores higher on coding benchmarks: 51.6 versus 45.2 in the Noometry coding category.
Which has the bigger context window?
Both accept 400K tokens.
How many benchmarks do GPT-5.2 and GPT-5.4 mini share?
42 benchmarks have published results for both models. GPT-5.2 has 67 scored results on Noometry and GPT-5.4 mini has 46.