BENCHMARK · 과학
GPQA Diamond 순위
대학원 수준의 생물·물리·화학 객관식 문제 198개. 검색으로 답을 찾기 어렵게 만든 문항이다.
상위 모델 여러 개가 95% 이상을 받아 이 시험은 최신 모델끼리의 차이를 가리기 어려워졌습니다(포화). 최신 모델 비교에는 다른 벤치마크를 함께 보세요.
상위 모델
- OpenAI
- Anthropic
Google- xAI
세로 점선은 무작위로 찍었을 때 기대되는 점수입니다.
공개일에 따른 점수
선은 개발사별로 그때까지의 최고 점수를 잇습니다. 오른쪽 위로 갈수록 최근에 나온 높은 점수입니다.
- GPT-6 Astra 96%
- Claude Sonnet 5.5 96%
- Gemini 3.8 Flash 95%
- Grok 4.6 94%
전체 순위
| 순위 | 모델 | 점수 | 생각 수준 | 공개일 |
|---|---|---|---|---|
| 1 | GPT-6 AstraOpenAI | 95.8%±1.4% | 최대 | 2026-09-03 |
| 2 | Claude Sonnet 5.5Anthropic | 95.6%±1.4% | 최대 | 2026-09-28 |
| 3 | GPT-6.1 SolOpenAI | 95.4%±1.4% | 최대 | 2026-09-29 |
| 3 | Gemini 3.8 FlashGoogle | 95.4%±1.4% | 높음 | 2026-09-02 |
| 5 | Gemini 3.7 FlashGoogle | 94.8%±1.4% | 높음 | 2026-08-13 |
| 6 | GPT-5.4 ProOpenAI | 94.6%±1.6% | 매우 높음 | 2026-03-05 |
| 7 | Gemini 3.1 ProGoogle | 94.4%±1.6% | — | 2026-02-19 |
| 8 | GPT-6 SolOpenAI | 94.3%±1.5% | 최대 | 2026-09-22 |
| 9 | Gemini 3.6 FlashGoogle | 94.1%±1.4% | — | 2026-07-21 |
| 10 | Grok 4.6xAI | 94%±1.5% | 높음 | 2026-08-12 |
| 10 | GPT-5.5OpenAI | 94%±1.6% | 매우 높음 | 2026-04-23 |
| 12 | GPT-5.5 ProOpenAI | 93.9%±1.6% | 매우 높음 | 2026-04-23 |
| 13 | Claude Opus 5Anthropic | 93.9%±1.5% | 최대 | 2026-07-24 |
| 14 | GPT-5.6 SolOpenAI | 93.5%±1.6% | 최대 | 2026-07-09 |
| 15 | Grok 4.5xAI | 93.4%±1.4% | 높음 | 2026-07-08 |
| 16 | GPT-5.6 TerraOpenAI | 93.3%±1.5% | 최대 | 2026-07-09 |
| 17 | GPT-5.4OpenAI | 93.3%±1.8% | 매우 높음 | 2026-03-05 |
| 18 | Gemini 3.5 FlashGoogle | 92.8%±1.6% | — | 2026-05-19 |
| 19 | Grok 4.7xAI | 92.7%±1.6% | 매우 높음 | 2026-09-21 |
| 20 | Gemini 3 ProGoogle | 92.6%±1.7% | — | 2025-11-18 |
| 21 | GPT-5.6 LunaOpenAI | 91.6%±1.7% | 최대 | 2026-07-09 |
| 22 | GPT-5.2OpenAI | 91.4%±1.8% | 매우 높음 | 2025-12-11 |
| 23 | Claude Opus 4.8Anthropic | 91%±1.9% | — | 2026-05-28 |
| 24 | Claude Opus 5.5Anthropic | 90.6%±1.9% | 최대 | 2026-09-22 |
| 25 | Claude Sonnet 5Anthropic | 90.5%±1.8% | 매우 높음 | 2026-06-30 |
| 25 | Claude Opus 4.6Anthropic | 90.5%±1.7% | — | 2026-02-05 |
| 27 | GPT-6 LunaOpenAI | 90.5%±1.8% | 최대 | 2026-09-22 |
| 28 | Claude Opus 4.7Anthropic | 90.1%±1.8% | 매우 높음 | 2026-04-16 |
| 29 | Claude Haiku 5.5Anthropic | 89.6%±2% | 최대 | 2026-10-07 |
| 30 | Gemini 3 FlashGoogle | 89.4%±2.2% | — | 2025-12-17 |
| 31 | Grok 4.20xAI | 89.3%±1.9% | — | 2026-02-17 |
| 32 | Grok 4.3 BetaxAI | 88.8%±2% | — | 2026-04-17 |
| 33 | GPT-5.1OpenAI | 87.6%±1.9% | 높음 | 2025-11-13 |
| 34 | Claude Sonnet 4.6Anthropic | 87.4%±2% | — | 2026-02-17 |
| 35 | Grok 4xAI | 87%±2% | — | 2025-07-09 |
| 36 | GPT-5.4 MiniOpenAI | 86.9%±2.4% | 매우 높음 | 2026-03-17 |
| 37 | GPT-5OpenAI | 86.2%±2.1% | 높음 | 2025-08-07 |
| 38 | Claude Opus 4.5Anthropic | 86.1%±2.1% | — | 2025-11-24 |
| 39 | Claude Fable 5Anthropic | 85.9%±2.5% | 최대 | 2026-06-09 |
| 40 | Gemini 2.5 Pro (Jun 2025)Google | 85.3%±2.1% | — | 2025-06-05 |
| 41 | Gemini 2.5 Pro (Mar 2025)Google | 83.8%±2.6% | — | 2025-03-25 |
| 42 | Gemini 3.5 Flash-LiteGoogle | 83.3%±2.7% | — | 2026-07-21 |
| 43 | GPT-5.5 InstantOpenAI | 82.5%±2.3% | — | 2026-05-05 |
| 44 | Claude Sonnet 4.5Anthropic | 82.3%±2.7% | — | 2025-09-29 |
| 45 | Gemini 3.1 Flash-LiteGoogle | 81.8%±2.8% | — | 2026-03-03 |
| 45 | o3OpenAI | 81.8%±2.1% | 높음 | 2025-04-16 |
| 47 | Claude 3.7 SonnetAnthropic | 79.7%±2.6% | — | 2025-02-24 |
| 48 | o4-miniOpenAI | 79.6%±2.4% | 높음 | 2025-04-16 |
| 49 | Claude Sonnet 4Anthropic | 79.2%±2.7% | — | 2025-05-22 |
| 50 | GPT-5.4 NanoOpenAI | 78.5%±2.4% | 높음 | 2026-03-17 |
| 51 | Claude Opus 4.1Anthropic | 77.3%±3% | — | 2025-08-05 |
| 52 | o3-miniOpenAI | 77%±2.6% | 높음 | 2025-01-31 |
| 53 | o1OpenAI | 76.8%±3% | 높음 | 2024-12-17 |
| 54 | Grok-3 minixAI | 76.3%±3% | — | 2025-06-24 |
| 54 | Claude Opus 4Anthropic | 76.3%±3% | — | 2025-05-22 |
| 56 | gpt-oss-120bOpenAI | 75.8%±2.7% | — | 2025-08-05 |
| 56 | Grok 3xAI | 75.8%±2.7% | — | 2025-04-09 |
| 58 | GPT-5 miniOpenAI | 75%±2.3% | 높음 | 2025-08-07 |
| 59 | Claude Haiku 4.5Anthropic | 71.2%±3.2% | — | 2025-10-15 |
| 60 | GPT-5 nanoOpenAI | 69.4%±2.8% | 높음 | 2025-08-07 |
| 61 | GPT-4.5OpenAI | 68.7%±3.3% | — | 2025-02-27 |
| 62 | GPT-4.1OpenAI | 66.9%±2.8% | — | 2025-04-14 |
| 63 | Gemini 2.5 Pro (May 2025)Google | 66.7%±3.4% | — | 2025-05-06 |
| 64 | GPT-4.1 miniOpenAI | 65.8%±2.7% | — | 2025-04-14 |
| 65 | Gemini 2.0 ProGoogle | 65.7%±3.4% | — | 2025-02-05 |
| 66 | o1-miniOpenAI | 62.4%±2.7% | 높음 | 2024-09-12 |
| 67 | gpt-oss-20bOpenAI | 60.8%±2.5% | — | 2025-08-05 |
| 68 | Gemini 1.5 Pro (Sept 2024)Google | 57.2%±2.8% | — | 2024-09-24 |
| 69 | Claude 3.5 Sonnet (October 2024)Anthropic | 55.3%±2.8% | — | 2024-10-22 |
| 70 | Claude 3.5 SonnetAnthropic | 54%±2.8% | — | 2024-06-20 |
| 71 | Grok-2 (Dec 2024)xAI | 53.8%±2.7% | — | 2024-12-12 |
| 72 | o1-previewOpenAI | 50.3%±2.8% | — | 2024-09-12 |
| 73 | GPT-4o (Aug 2024)OpenAI | 49.2%±2.6% | — | 2024-08-06 |
| 74 | GPT-4.1 nanoOpenAI | 48.9%±2.5% | — | 2025-04-14 |
| 75 | GPT-4o (May 2024)OpenAI | 48.9%±2.6% | — | 2024-05-13 |
| 76 | GPT-4o (Nov 2024)OpenAI | 47.9%±2.6% | — | 2024-11-20 |
| 77 | Gemini 1.5 Flash (Sep 2024)Google | 47.3%±2.8% | — | 2024-09-24 |
| 78 | Claude 3 OpusAnthropic | 47.2%±2.6% | — | 2024-02-29 |
| 79 | GPT-4 Turbo (Apr 2024)OpenAI | 46.6%±2.7% | — | 2024-04-09 |
| 80 | Gemini 1.5 Pro (May 2024)Google | 45.9%±2.6% | — | 2024-05-14 |
| 81 | GPT-4 Turbo (Nov 2023)OpenAI | 42.4%±2.4% | — | 2024-01-25 |
| 82 | Claude 3 SonnetAnthropic | 40.6%±2.4% | — | 2024-02-29 |
| 83 | Gemini 1.5 Flash (May 2024)Google | 40.4%±2.4% | — | 2024-05-23 |
| 84 | Claude 3.5 HaikuAnthropic | 38.1%±2.6% | — | 2024-10-22 |
| 85 | GPT-4o miniOpenAI | 37.7%±2.4% | — | 2024-07-18 |
| 86 | Claude 3 HaikuAnthropic | 36.3%±2.4% | — | 2024-03-07 |
| 87 | GPT-4 (Mar 2023)OpenAI | 35.7%±2.4% | — | 2023-03-14 |
| 88 | Claude 2Anthropic | 34.7%±2.5% | — | 2023-07-11 |
| 89 | Gemini 1.0 ProGoogle | 34%±2% | — | 2023-12-13 |
| 90 | Gemini 1.5 Flash 8BGoogle | 33%±2.4% | — | 2024-10-03 |
| 90 | Claude 2.1Anthropic | 33%±2.3% | — | 2023-11-21 |
| 92 | GPT-4 (Jun 2023)OpenAI | 30.6%±2.3% | — | 2023-06-13 |
| 93 | GPT-3.5 Turbo (Nov 2023)OpenAI | 28%±1.9% | — | 2023-11-06 |
| 94 | GPT-3.5 Turbo (Jan 2024)OpenAI | 27.2%±1.9% | — | 2024-01-25 |
다른 벤치마크
- MATH Level 5 수학
- OTIS 모의 AIME 수학
- FrontierMath(2025판) 수학
- FrontierMath 1~3단계 수학
- FrontierMath 4단계 수학
- SWE-bench Verified 코딩
- MirrorCode 코딩
- SimpleQA Verified 지식
- 체스 퍼즐 추론
- 미공개 게임 퍼즐 추론
- 가구 조립 검사 공간·학습
- EBR-bench 공간·학습
- Aider Polyglot 코딩