GPQA Diamond

Name: GPQA Diamond
Creator: aib vote
License: https://www.gnu.org/licenses/gpl-3.0.html

About This Benchmark

A multiple-choice benchmark measuring graduate-level expert knowledge in biology, chemistry, and physics — challenging even for domain experts. Score is accuracy (%).

Source: Artificial Analysis

Rank	Model
#1	Google Gemini 3.1 Pro	94.1%
#2	OpenAI GPT-5.4	92.0%
#3	Anthropic Claude Opus 4.7	91.4%
#4	Grok Grok 4.20 (Reasoning)	91.1%
#5	Google Gemini 3 Flash	89.8%
#6	Anthropic Claude Opus 4.6	89.6%
#7	DeepSeek DeepSeek V4 Flash	89.4%
#8	DeepSeek DeepSeek V4 Pro	88.8%
#9	Meta Muse Spark	88.4%
#10	Alibaba Qwen3.6 Plus	88.2%
#11	Moonshot AI Kimi K2.5	87.9%
#12	MiniMax MiniMax M2.7	87.4%
#13	Xiaomi MiMo-V2-Pro	87.0%
#14	Z.ai GLM-5.1	86.8%
#15	Anthropic Claude Opus 4.5	86.6%
#16	Alibaba Qwen3.5 397B A17B	86.1%
#17	Google Gemma 4 31B	85.7%
#18	Grok Grok 4.1 Fast (Reasoning)	85.3%
#19	MiniMax MiniMax M2.5	84.8%
#20	Google Gemini 2.5 Pro	84.4%
#21	DeepSeek DeepSeek V3.2	84.0%
#22	Anthropic Claude Sonnet 4.5	83.4%
#23	OpenAI GPT-5.4 Mini	82.3%
#24	Google Gemini 3.1 Flash Lite	82.2%
#25	Z.ai GLM-5	82.0%
#26	Baidu ERNIE 4.5 300B A47B	81.1%
#27	Anthropic Claude Opus 4.1	80.9%
#28	NVIDIA Nemotron 3 Super	80.0%
#29	Anthropic Claude Sonnet 4.6	79.7%
#30	Anthropic Claude Opus 4	79.6%
#31	Google Gemini 2.5 Flash	79.0%
#32	Anthropic Claude Sonnet 4	77.7%
#33	Baidu ERNIE 5.0 Thinking	77.7%
#34	Grok Grok 4.20	77.6%
#35	Mistral AI Mistral Small 4	76.9%
#36	OpenAI GPT-5.4 Nano	76.1%
#37	Arcee AI Trinity Large Thinking	75.2%
#38	OpenAI GPT-5 Mini	68.7%
#39	OpenAI GPT-5	68.6%
#40	Anthropic Claude Haiku 4.5	67.2%
#41	OpenAI GPT OSS 120B	67.2%
#42	Meta Llama 4 Maverick	67.1%
#43	OpenAI GPT-5 Nano	67.0%
#44	OpenAI GPT-4.1	66.6%
#45	Google Gemini 2.5 Flash Lite	65.1%
#46	Grok Grok 4.1 Fast	63.7%
#47	Meituan Longcat Flash Chat	63.6%
#48	Amazon Nova 2 Lite	60.3%
#49	Meta Llama 4 Scout	58.7%