GPT-4.1 mini provides a balance between intelligence, speed, and cost. It's a significant leap in small model performance, even beating GPT-4o in many benchmarks while reducing latency and cost.
发布日期2025年4月14日
参数规模—
上下文长度1.0M
许可证Proprietary
知识截止2024年5月31日
Benchmarks
评测成绩
| 评测基准 | 类别 | 分数 | 来源 |
|---|---|---|---|
| 安全防御榜·Bias | security | 95.3 | — |
| 安全防御榜·Toxicity | security | 91.7 | — |
| 安全防御榜·Info Safety | security | 90.0 | — |
| 多模态安全榜·Legality | security vision | 88.9 | — |
| 多模态安全榜·Health Risk | security vision | 88.7 | — |
| CharXiv-D | reasoning vision multimodal | 88.4 | 来源 |
| 安全防御榜 | security | 88.1 | — |
| MMLU | general reasoning language math | 87.5 | 来源 |
| 安全防御榜·Malicious Use | security | 87.3 | — |
| 多模态安全榜·Toxicity | security vision | 87.0 | — |
| 多模态安全榜·Misinformation | security vision | 86.2 | — |
| 多模态安全榜 | security vision | 85.7 | — |
| 安全防御榜·Human Rights | security | 85.4 | — |
| 安全防御榜·Socioeconomic | security | 84.9 | — |
| 多模态安全榜·Bias | security vision | 84.4 | — |
| IFEval | general | 84.1 | 来源 |
| 多模态开源模型榜·OCRBench | vision | 84.0 | 来源 |
| 安全防御榜·Child Safety | security | 81.9 | — |
| 多模态开源模型榜·MMBench TEST | vision | 80.9 | 来源 |
| 多模态安全榜·Privacy | security vision | 79.0 | — |
| MMMLU | language reasoning math general | 78.5 | 来源 |
| 多模态开源模型榜·AI2D | vision | 76.0 | 来源 |
| 多模态开源模型榜·MMVet | vision | 74.3 | 来源 |
| MathVista | math vision multimodal | 73.1 | 来源 |
| MMMU | multimodal reasoning general | 72.7 | 来源 |
| 多模态开源模型榜·MathVista | vision | 70.9 | 来源 |
| 多模态开源模型榜 | vision | 68.9 | 来源 |
| Multi-IF | reasoning communication language | 67.0 | 来源 |
| GPQA | reasoning general | 65.0 | 来源 |
| Graphwalks BFS <128k | reasoning spatial_reasoning | 61.7 | 来源 |
| 多模态开源模型榜·MMStar | vision | 60.9 | 来源 |
| Graphwalks parents <128k | reasoning spatial_reasoning | 60.5 | 来源 |
| 价值对齐榜·Fairness | security | 59.8 | — |
| CharXiv-R | reasoning vision multimodal | 56.8 | 来源 |
| TAU-bench Retail | reasoning communication | 55.8 | 来源 |
| 多模态开源模型榜·MMMU VAL | vision | 55.0 | 来源 |
| COLLIE | language reasoning writing | 54.6 | 来源 |
| AIME 2024 | math reasoning | 49.6 | 来源 |
| 多模态开源模型榜·HallusionBench | vision | 49.3 | 来源 |
| ComplexFuncBench | long_context reasoning | 49.3 | 来源 |
| OpenAI-MRCR: 2 needle 128k | long_context reasoning | 47.2 | 来源 |
| 价值对齐榜·Legality | security | 45.2 | — |
| Internal API instruction following (hard) | general | 45.1 | 来源 |
| 价值对齐榜·Data Protection | security | 44.6 | — |
| 价值对齐榜 | security | 43.9 | — |
| MultiChallenge (o3-mini grader) | reasoning language | 42.2 | 来源 |
| AIME 2025 | math reasoning | 40.2 | 来源 |
| TAU-bench Airline | reasoning communication | 36.0 | 来源 |
| 价值对齐榜·Morality | security | 35.9 | — |
| MultiChallenge | communication reasoning | 35.8 | 来源 |
| HMMT 2025 | math | 35.0 | 来源 |
| Aider-Polyglot | general code | 34.7 | 来源 |
| 价值对齐榜·Safety | security | 33.9 | — |
| OpenAI-MRCR: 2 needle 1M | long_context reasoning | 33.3 | 来源 |
| Aider-Polyglot Edit | general code | 31.6 | 来源 |
| SWE-Bench Verified | reasoning frontend_development code | 23.6 | 来源 |
| Graphwalks BFS >128k | reasoning spatial_reasoning long_context | 15.0 | 来源 |
| Graphwalks parents >128k | reasoning spatial_reasoning long_context | 11.0 | 来源 |
| Humanity's Last Exam | general | 3.7 | 来源 |
Pricing
API 价格对比
| 服务商 | 输入价 | 输出价 | 上下文 | 吞吐(tok/s) | 延迟(s) | 函数调用 | 代码执行 | 联网搜索 |
|---|---|---|---|---|---|---|---|---|
| 302.AI | $0.40 | $1.60 | 1.0M | — | — | ✓ | ✗ | ✗ |
| Abacus | $0.40 | $1.60 | 1.0M | — | — | ✓ | ✗ | ✗ |
| Azure | $0.40 | $1.60 | 1.0M | — | — | ✓ | ✗ | ✗ |
| Azure Cognitive Services | $0.40 | $1.60 | 1.0M | — | — | ✓ | ✗ | ✗ |
| Helicone | $0.40 | $1.60 | 1.0M | — | — | ✓ | ✗ | ✗ |
| Helicone | $0.40 | $1.60 | 1.0M | — | — | ✓ | ✗ | ✗ |
| LLM Gateway | $0.40 | $1.60 | 1.0M | — | — | ✓ | ✗ | ✗ |
| OpenAI | $0.40 | $1.60 | 1.0M | — | — | ✓ | ✗ | ✗ |
| Pioneer | $0.40 | $1.60 | 1.0M | — | — | ✓ | ✗ | ✗ |
| SAP AI Core | $0.40 | $1.60 | 1.0M | — | — | ✓ | ✗ | ✗ |
价格单位：美元/百万 token，数据来自社区整理，仅供参考。