GPT-4.5 is OpenAI's most advanced model, offering improved reasoning, coding, and creative capabilities with faster performance and longer context handling than GPT-4. It features enhanced instruction following, reduced hallucinations, and better factual accuracy.
发布日期2025年2月27日
参数规模—
上下文长度—
许可证Proprietary
知识截止—
Benchmarks
评测成绩
| 评测基准 | 类别 | 分数 | 来源 |
|---|---|---|---|
| GSM8k | math reasoning | 97.0 | 来源 |
| MMLU | general reasoning language math | 90.8 | 来源 |
| CharXiv-D | reasoning vision multimodal | 90.0 | 来源 |
| IFEval | general | 88.2 | 来源 |
| HumanEval | reasoning code | 88.0 | 来源 |
| MMMLU | language reasoning math general | 85.1 | 来源 |
| MMMU | multimodal reasoning general | 75.2 | 来源 |
| Graphwalks parents <128k | reasoning spatial_reasoning | 72.6 | 来源 |
| COLLIE | language reasoning writing | 72.3 | 来源 |
| Graphwalks BFS <128k | reasoning spatial_reasoning | 72.3 | 来源 |
| MathVista | math vision multimodal | 72.3 | 来源 |
| Multi-IF | reasoning communication language | 70.8 | 来源 |
| GPQA | reasoning general | 69.5 | 来源 |
| TAU-bench Retail | reasoning communication | 68.4 | 来源 |
| ComplexFuncBench | long_context reasoning | 63.0 | 来源 |
| SimpleQA | general reasoning | 62.5 | 来源 |
| CharXiv-R | reasoning vision multimodal | 55.4 | 来源 |
| Internal API instruction following (hard) | general | 54.0 | 来源 |
| MultiChallenge (o3-mini grader) | reasoning language | 50.1 | 来源 |
| TAU-bench Airline | reasoning communication | 50.0 | 来源 |
| Aider-Polyglot Edit | general code | 44.9 | 来源 |
| MultiChallenge | communication reasoning | 43.8 | 来源 |
| OpenAI-MRCR: 2 needle 128k | long_context reasoning | 38.5 | 来源 |
| SWE-Bench Verified | reasoning frontend_development code | 38.0 | 来源 |
| SWE-Lancer | reasoning code | 37.3 | 来源 |
| AIME 2024 | math reasoning | 36.7 | 来源 |
| SWE-Lancer (IC-Diamond subset) | reasoning code | 17.4 | 来源 |
Pricing
API 价格对比
暂无 API 价格。