Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

SimpleBench

A multiple-choice text benchmark designed to test basic reasoning capabilities where non-specialized humans (high school level) consistently outperform state-of-the-art language models, covering spatio-temporal reasoning, social intelligence, and linguistic adversarial robustness.
Source:

Model Performance

#1
Gemini 3.1 Pro Preview
79.6%
#2
Claude Opus 4.6 (Thinking)
67.6%
#3
Claude Opus 4.6
67.6%
#4
Gemini 2.5 Pro (Jun 2025)
62.4%
#5
Claude Opus 4.5
62.0%
#6
Claude 4.1 Opus
60.0%
#7
Claude 4.1 Opus (Thinking)
60.0%
#8
Claude 4.0 Opus
58.8%
#9
Claude 4.0 Opus (Thinking)
58.8%
#10
GPT-5 (Thinking)
56.7%
#11
GPT-5
56.7%
#12
Claude Sonnet 4.5
54.3%
#13
OpenAI o3 (High Effort)
53.1%
#14
Gemini 2.5 Pro
51.6%
#15
Gemini 2.5 Pro (Thinking)
51.6%
#16
Claude 3.7 Sonnet (Thinking)
46.4%
#17
GPT-5.2
45.8%
#18
Claude 4.0 Sonnet (Thinking)
45.5%
#19
Claude 3.7 Sonnet
44.9%
#20
Claude 3.5 Sonnet
41.4%
#21
OpenAI o1
40.1%
#22
OpenAI o4 Mini (High Effort)
38.7%
#23
GPT-4.5
34.5%
#24
Qwen 3
31.0%
#25
Qwen 3 (Thinking)
31.0%
#26
DeepSeek R1
30.9%
#27
Gemini 2.0 Flash Thinking (Jan 2025)
30.7%
#28
Llama 4 Maverick 17B
27.7%
#29
DeepSeek V3 (Mar 2025)
27.2%
#30
Gemini 1.5 Pro
27.1%
#31
GPT-4.1
27.0%
#32
GPT-4o
25.1%
#33
Claude 3 Opus
23.5%
#34
Llama 3.1 405B
23.0%
#35
OpenAI o3 Mini (High Effort)
22.8%
#36
Mistral Large 2
22.5%
#37
GPT OSS 120B
22.1%
#38
Mistral Large 3
20.4%
#39
Llama 3.3 70B
19.9%
#40
DeepSeek V3
18.9%
#41
Gemini 2.0 Flash
18.9%
#42
OpenAI o1 Mini
18.1%
#43
Cohere Command R+
17.4%
#44
GPT-4o Mini
10.7%