GPQA leaderboard

A challenging dataset of 448 multiple-choice questions written by domain experts in biology, physics, and chemistry. Designed to be difficult even for skilled non-experts with access to Google.

Models ranked by their published Graduate-Level Google-Proof Q&A value

Rank	Model	Provider	Multiple choice questions
1	GPT-6 Astra	OpenAI	96
2	SASakana Fugu	Sakana AI	95.5
2	SASakana Fugu-Ultra	Sakana AI	95.5
4	GPT-5.6 Sol	OpenAI	94.6
5	Claude Opus 4.7 (Adaptive)	Anthropic	94.2
6	Claude Mythos 5	Anthropic	94.1
7	Claude Opus 4.8	Anthropic	93.6
7	GPT-5.5	OpenAI	93.6
9	Kimi K3	Moonshot AI	93.5
10	GPT-5.6 Terra	OpenAI	92.9
11	GPT-5.4	OpenAI	92.8
11	OAOrnith-1.5-397B	Ornith AI	92.8
13	Qwen3.8 Max	Alibaba	92.6
14	GPT-5.2	OpenAI	92.4
14	Qwen3.7 Max	Alibaba	92.4
16	GPT-5.6 Luna	OpenAI	92.3
16	Hy4 preview	Tencent	92.3
18	Gemini 3.5 Flash	Google	92.2
19	Qwen3.8-Flash-Next	Alibaba	91.7
20	Claude Opus 4.6	Anthropic	91.3
21	GLM-5.2	Z.AI	91.2
22	Qwen3.8-Omni-Flash	Alibaba	91
23	DeepSeek V4.1 Flash	DeepSeek	90.9
24	Kimi K2.6	Moonshot AI	90.5
25	Qwen3.6 Plus	Alibaba	90.4
26	Qwen3.7 Plus	Alibaba	90.3
27	DeepSeek V4 Pro 0813	DeepSeek	90.1
27	Grok 4.3	xAI	90.1
29	Claude Sonnet 4.6	Anthropic	89.9
29	INInterfaze Beta	Interfaze	89.9
31	TMInkling-Small	Thinking Machines Lab	89.5
32	OAOrnith-1.5-35B-A3B	Ornith AI	89.2
32	Qwen3.8-27B	Alibaba	89.2
34	Qwen3.5 397B	Alibaba	88.4
35	DeepSeek V4 Flash 0731	DeepSeek	88.1
36	GPT-5.4 mini	OpenAI	88
37	TMInkling	Thinking Machines Lab	87.9
38	Qwen3.6-27B	Alibaba	87.8
39	Kimi K2.5	Moonshot AI	87.6
39	Kimi K2.5 (Reasoning)	Moonshot AI	87.6
41	Hy3 Preview	Tencent	87.2
42	Claude Opus 4.5	Anthropic	87
42	Nemotron 3 Ultra	NVIDIA	87
44	Qwen3.5-122B-A10B	Alibaba	86.6
45	OAOrnith-1.5-9B	Ornith AI	86.4
46	GLM-5	Z.AI	86
46	Qwen3.6-35B-A3B	Alibaba	86
48	PMTernary Bonsai 2 27B	Prism ML	85.76
49	GLM-4.7	Z.AI	85.7
50	Qwen3.5-27B	Alibaba	85.5
51	Ling 3.0 Flash	InclusionAI	84.97
52	Gemma 4 31B	Google	84.3
53	MAI-Thinking-1	Microsoft	84.2
53	Qwen3.5-35B-A3B	Alibaba	84.2
55	Ling 3.0 Flash FP8	InclusionAI	84
56	MiMo-V2-Flash	Xiaomi	83.7
57	Claude Sonnet 4.5	Anthropic	83.4
58	Gemini 2.5 Pro	Google	83
59	GPT-5.4 nano	OpenAI	82.8
60	o1-pro	OpenAI	79
61	Gemma 4 12B	Google	78.8
62	Qwen3 235B 2507	Alibaba	77.5
63	o3-mini	OpenAI	77.2
64	o1	OpenAI	75.7
65	Nemotron 3.5 Lightning 30B A3B NVFP4	NVIDIA	75.57
66	Nemotron 3 Nano Omni 30B A3B	NVIDIA	72.2
67	ZYZAYA1-8B	Zyphra	71
68	Granite 4.2 30B	IBM	66.41
69	GPT-4.1	OpenAI	66.3
70	GPT-4.1 mini	OpenAI	64.2
71	Granite 4.2 8B	IBM	64.14
72	Claude 3.5 Sonnet	Anthropic	59.4
73	DeepSeek V3	DeepSeek	59.1
74	Ling 2.6 Flash	InclusionAI	59
75	Gemma 4 E4B	Google	58.6
76	JEMellum2-12B-A2.5B-Thinking	JetBrains	57.6
77	ZYZAYA1-74B-Preview	Zyphra	57.3
78	Granite 4.2 3B	IBM	54.8
79	GPT-4.1 nano	OpenAI	50.3
80	Gemma 4 E2B	Google	43.4
80	SPSoofi S 30B-A3B	Soofi Project	43.4
82	JEMellum2-12B-A2.5B-Instruct	JetBrains	40.9
83	LFM2.5-VL-450M	LiquidAI	25.66
84	LFM2.5-230M	LiquidAI	25.41

Evidence key: Observed

Rows are ordered by the value GPQA: A Graduate-Level Google-Proof Q&A Benchmark published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.