Benchmark Performance
| Benchmark | Score | Unit | Evaluated At | Notes | Source |
|---|
| MMLU | 85.8 | % | 2023-03-14 | 5-shot | view |
| HUMANEVAL | 77.6 | pass@1 | 2023-03-14 | — | view |
| GSM8K | 91.7 | % | 2023-03-14 | 0-shot CoT | view |
| MATH | 43.9 | % | 2023-03-14 | 0-shot CoT | view |
| BBH | 80.9 | % | 2023-03-14 | 3-shot CoT | view |
| GPQA | 39.2 | % | 2023-03-14 | 0-shot | view |
| IFEVAL | 74.6 | % | 2023-03-14 | prompt_strict | view |
| ARC | 94.1 | % | 2023-03-14 | challenge | view |
| MUSR | 54.6 | % | 2023-03-14 | 0-shot | view |
| WINOGRANDE | 80.3 | % | 2023-03-14 | 0-shot | view |
Pricing
| Tier | Price | Currency |
|---|
| Input | $30 / Mtok | USD |
| Output | $60 / Mtok | USD |
| Cache Read | $0 / Mtok | USD |
| Cache Write | $0 / Mtok | USD |
Source:
https://openai.com/api/pricing/
· as of 2023-03-14
Compliance
- Data Residency: US
- SOC2: ✓
- HIPAA: ✓
- GDPR: ✓
- ISO 27001: ✓
GPT-4
모델 개요
OpenAI GPT-4 大型语言模型, 8K 上下文, 在推理/创意/协作任务上显著超越 GPT-3.5, 首次引入人类水平表现。
핵심 사양
| 공급업체 | 버전 | 출시일 | 컨텍스트 창 | 입력 모달리티 | 출력 모달리티 | 라이선스 |
|---|
| Openai | 4 | 2023-03-14 | 8K | text | text | Proprietary |
벤치마크 성능
| 벤치마크 | 점수 | 단위 | 비고 |
|---|
| MMLU (Massive Multitask Language Understanding) | 85.8 | % | 5-shot |
| HumanEval | 77.6 | pass@1 | — |
| GSM8K (Grade School Math 8K) | 91.7 | % | 0-shot CoT |
| MATH | 43.9 | % | 0-shot CoT |
| BBH (BIG-Bench Hard) | 80.9 | % | 3-shot CoT |
| GPQA | 39.2 | % | 0-shot |
| IFEval | 74.6 | % | prompt_strict |
| ARC | 94.1 | % | challenge |
| MUSR | 54.6 | % | 0-shot |
| WinoGrande | 80.3 | % | 0-shot |
가격
백만 토큰당
강점
- MMLU score 85.8, strong knowledge reasoning.
- GSM8K 91.7, robust math reasoning.
약점
- 闭源专有模型，不支持自托管。
- 上下文窗口 8K 偏小。
사용 사례
참고문헌