GPQA Diamond

GPQA에서 가장 어려운 198개 문항으로, 박사급 전문가는 65% 정확도를 달성하지만 숙련된 비전문가는 웹 접근에도 불구하고 34%에 그칩니다.

모델: 205 현재, 617 전체
중앙값: 71.3%
범위: 11.1%–96.3%

점수 분포

가장 낮음: 11.1%
중앙값: 71.3%
최고: 96.3%

리더보드

지원 중단된 모델 포함

순위	모델	점수
51	o3 OpenAI	82.7%
52	Qwen3.5 122B A10B (Non-reasoning) Alibaba	82.7%
53	Qwen3.5 Omni Plus Alibaba	82.6%
54	GPT-5.5 Instant (June 2026) OpenAI	82.3%
55	K2 Horizon MoVA 36B A4B Institute of Foundation Models	82.2%
56	Qwen3.8 27B (Non-reasoning) Alibaba	81.8%
57	Qwen3.6 35B A3B (Non-reasoning) Alibaba	81.7%
58	ERNIE 4.5 300B A47B Baidu	81.1%
59	Nova 2.0 Lite (High) Amazon	81.1%
60	Qwen3.5 9B (Reasoning) Alibaba	80.6%
61	G9v3-39A5B AI9Stars	80.5%
62	EXAONE 4.5 33B (Reasoning) LG AI Research	79.4%
63	Gemma 4 26B A4B (Reasoning) Google	79.2%
64	Qwen3.5 9B (Non-reasoning) Alibaba	78.6%
65	Nova 2.0 Pro Preview (Medium) Amazon	78.5%
66	gpt-oss-120b (High) OpenAI	78.2%
67	LongCat 2.0 LongCat	78.0%
68	ERNIE 5.0 Thinking Preview Baidu	77.7%
69	Qwen3.5 4B (Reasoning) Alibaba	77.1%
70	Mistral Small 4 (Reasoning) Mistral	76.9%
71	Cogito v2.1 (Reasoning) Deep Cogito	76.8%
72	Nova 2.0 Lite (Medium) Amazon	76.8%
73	Doubao Seed Code ByteDance Seed	76.4%
74	Gemma 4 31B (Non-reasoning) Google	76.3%
75	Command A+ Cohere	76.1%
76	INTELLECT-3 (Based on GLM-4.5-Air) Prime Intellect	76.1%
77	Nova 2.0 Omni (Medium) Amazon	76.0%
78	Qwen3 Next 80B A3B (Reasoning) Alibaba	75.9%
79	Nemotron Cascade 2 30B A3B NVIDIA	75.8%
80	North Mini Code Cohere	75.7%
81	NVIDIA Nemotron 3 Nano 30B A3B (Reasoning) NVIDIA	75.7%
82	Gemma 4 12B (Reasoning) Google	75.3%
83	K2 Horizon 7B Institute of Foundation Models	75.3%
84	Trinity Large Thinking Arcee AI	75.2%
85	Nova 2.0 Pro Preview (Low) Amazon	75.1%
86	Mistral Medium 3.5 Mistral	74.8%
87	Nemotron 3.5 Lightning NVIDIA	74.3%
88	Qwen3.5 Omni Flash Alibaba	74.2%
89	EXAONE 4.0 32B (Reasoning) LG AI Research	73.9%
90	Magistral Medium 1.2 Mistral	73.9%
91	Qwen3 Next 80B A3B Instruct Alibaba	73.8%
92	Sarvam 105B (High) Sarvam	73.8%
93	Qwen3 Coder Next Alibaba	73.7%
94	Ling 3.0 Tiny InclusionAI	73.4%
95	Apriel-v1.6-15B-Thinker ServiceNow	73.3%
96	HyperNova 60B 2605 (High, Based on gpt-oss-120b) Multiverse Computing	73.3%
97	Quasar 438B (Max, Based on GLM-5.2) Multiverse Computing	73.2%
98	Llama 3.1 Nemotron Ultra 253B v1 (Reasoning) NVIDIA	72.8%
99	Hermes 4 - Llama-3.1 405B (Reasoning) Nous Research	72.7%
100	Qwen3 Omni 30B A3B (Reasoning) Alibaba	72.6%

51
o3
OpenAI
82.7%
52
Qwen3.5 122B A10B (Non-reasoning)

데이터 출처: Artificial Analysis

Command Palette

GPQA Diamond

점수 분포

리더보드