Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

GPQA

Measures accuracy on the most challenging subset of the Google-Proof Question Answering benchmark, which includes difficult, expert-level questions.
Source:

Model Performance

#1
Grok-4-Thinking
84.1%
#2
Grok-4
84.1%
#3
OpenAI-O3-medium
78.1%
#4
OpenAI-O3-high
78.1%
#5
Gemini-2.5-Pro-Thinking
73.7%
#6
Gemini-2.5-Pro-0325
73.7%
#7
GPT-5-mini
73.7%
#8
Grok-3-Mini-Beta
72.0%
#9
Qwen-3-Max-Preview
70.4%
#10
Claude-4.1-Opus-Thinking
67.3%
#11
Claude-3.7-Sonnet-Thinking
67.1%
#12
OpenAI-O3-mini-high
66.7%
#13
OpenAI-O3-mini-medium
66.7%
#14
Claude-4.0-Sonnet-Thinking
66.0%
#15
OpenAI-O4-mini-high
66.0%
#16
OpenAI-O4-mini-medium
66.0%
#17
Grok-3-Beta
64.9%
#18
Grok-3-Think
64.9%
#19
OpenAI-O1-1217
64.0%
#20
Claude-4.0-Opus
62.3%
#21
Kimi-K2-Instruct
62.0%
#22
Claude-4.1-Opus
59.9%
#23
Claude-4.0-Sonnet
59.2%
#24
Llama-4-Maverick-17B
56.9%
#25
Claude-3.7-Sonnet
56.5%
#26
Qwen-3
55.2%
#27
Qwen-3-Thinking
55.2%
#28
Gemini-2.0-Flash
53.6%
#29
GPT-4.1
52.8%
#30
DeepSeek-V3-0324
48.1%
#31
Claude-3.5-Sonnet-1022
45.5%
#32
Gemini-2.5-Flash-Thinking
44.8%
#33
Gemini-1.5-Pro
44.4%
#34
DeepSeek-V3
38.7%
#35
Gemini-2.5-Flash
37.7%
#36
GPT-4o-0513
33.7%
#37
Llama-3.3-70B
33.3%
#38
Mistral-Large-2
26.9%
#39
GPT-4-mini
25.6%
#40
Claude-3.5-Haiku
17.2%
#41
Cohere-Command-A
5.7%
#42
GPT-3.5-Turbo
5.7%
#43
Cohere-Command-R-Plus
5.3%