GPT-5.2 vs GPT-5.4
Côte à côte. Chaque métrique. Chaque benchmark.
| Type | GPT-5.2 | GPT-5.4 |
|---|---|---|
| Provider |  OpenAI | OpenAI |
| score moyen | 54.0 | 59.0 |
| Prix d'entrée | $1.75 | $2.50 |
| Prix de sortie | $14.00 | $15.00 |
| Fenêtre de contexte | 400K tokens (~200 books) | 1.1M tokens (~525 books) |
| Sorti le | 2025-12-10 | 2026-03-05 |
| Code source ouvert | Proprietary | Proprietary |
Scores de benchmark
13 benchmarks · GPT-5.2: 4, GPT-5.4: 9
| Benchmark | Catégorie | GPT-5.2 | GPT-5.4 |
|---|---|---|---|
| APEX-Agents | agentic | 34.3 | 35.9 |
| ARC-AGI | reasoning | 86.2 | 93.7 |
| ARC-AGI-2 | reasoning | 52.9 | 74.0 |
| Chatbot Arena Elo — Overall | arena | 1439.5 | 1465.8 |
| Chess Puzzles | knowledge | 49.0 | 44.0 |
| FrontierMath-2025-02-28-Private | math | 40.7 | 47.6 |
| FrontierMath-Tier-4-2025-07-01-Private | math | 18.8 | 27.1 |
| GPQA diamond | knowledge | 88.5 | 91.1 |
| OTIS Mock AIME 2024-2025 | math | 96.1 | 95.3 |
| PostTrainBench | knowledge | 21.4 | 20.2 |
| SimpleQA Verified | knowledge | 38.9 | 44.8 |
| SWE-Bench verified | coding | 73.8 | 76.9 |
| WeirdML | coding | 72.2 | 57.4 |