Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

Humanity's Last Exam

A comprehensive benchmark developed by the Center for AI Safety and Scale AI to assess expert-level reasoning and knowledge across diverse fields. The dataset contains 3,000 questions from nearly 1,000 subject-matter experts across 500+ institutions and 50 countries, with 10% requiring image and text comprehension.
Source:

Model Performance

#1
Gemini 3.1 Pro Preview
45.0%
#2
GPT-5.3 Codex
40.0%
#3
Claude Opus 4.6 (Thinking)
37.0%
#4
GPT-5.2 Codex
34.0%
#5
Grok 4.20 (Reasoning)
30.0%
#6
Claude Opus 4.5 (Thinking)
28.0%
#7
MiniMax M2.7
28.0%
#8
GPT-5 (Thinking)
27.0%
#9
GPT-5
27.0%
#10
Grok 4 (Thinking)
24.0%
#11
Grok 4
24.0%
#12
Grok 4.20
23.0%
#13
GPT-5.1 Codex Max
23.0%
#14
MiniMax M2.1
22.0%
#15
DeepSeek V3.2 (Thinking)
22.0%
#16
Gemini 2.5 Pro (Thinking)
21.0%
#17
OpenAI o3 (High Effort)
20.0%
#18
OpenAI o3 (Medium Effort)
20.0%
#19
GPT-5 Mini (Thinking)
20.0%
#20
Claude Opus 4.6
19.0%
#21
MiniMax M2.5
19.0%
#22
GPT OSS 120B
19.0%
#23
OpenAI o4 Mini (High Effort)
18.0%
#24
OpenAI o4 Mini (Medium Effort)
18.0%
#25
Claude Sonnet 4.5 (Thinking)
17.0%
#26
Gemini 2.5 Pro
17.0%
#27
Gemini 3.1 Flash Lite Preview
16.0%
#28
Qwen 3 (Thinking)
15.0%
#29
Gemini 2.5 Pro (Jun 2025)
15.0%
#30
DeepSeek R1
15.0%
#31
Gemini 3.0 Flash
14.0%
#32
Claude Opus 4.5
13.0%
#33
Claude Sonnet 4.6
13.0%
#34
Qwen 3 Max (Thinking)
12.0%
#35
OpenAI o3 Mini (High Effort)
12.0%
#36
Kimi K2.5
12.0%
#37
Claude 4.0 Opus (Thinking)
12.0%
#38
Claude 4.1 Opus (Thinking)
12.0%
#39
Grok 3 Mini
11.0%
#40
DeepSeek V3.2
11.0%
#41
Qwen 3
11.0%
#42
Gemini 2.5 Flash (Thinking)
11.0%
#43
GPT-5.4
11.0%
#44
Magistral Medium 3.1
10.0%
#45
Claude 3.7 Sonnet (Thinking)
10.0%
#46
Claude Haiku 4.5 (Thinking)
10.0%
#47
Claude 4.0 Sonnet (Thinking)
10.0%
#48
Qwen 3 Max Preview
9.0%
#49
OpenAI o3 Mini (Medium Effort)
9.0%
#50
GPT-5 Nano (Thinking)
8.0%
#51
OpenAI o1
8.0%
#52
MiMo V2 Flash
8.0%
#53
Claude Sonnet 4.5
7.0%
#54
GLM-5
7.0%
#55
Gemini 2.0 Pro
7.0%
#56
Kimi K2
7.0%
#57
Gemini 2.0 Flash Thinking (Jan 2025)
7.0%
#58
GPT-5.2
7.0%
#59
Llama 2 7B
6.0%
#60
GPT-5.4 Mini
6.0%
#61
Claude 4.0 Opus
6.0%
#62
GLM-4.7
6.0%
#63
DeepSeek V3 (Mar 2025)
5.0%
#64
Cohere Command A
5.0%
#65
OpenAI o1 Mini
5.0%
#66
Llama 2 13B
5.0%
#67
Llama 2 70B
5.0%
#68
Grok 3
5.0%
#69
GPT-4.1
5.0%
#70
Llama 4 Maverick 17B
5.0%
#71
Gemini 1.5 Pro
5.0%
#72
Gemini 2.5 Flash
5.0%
#73
GPT-5 Mini
5.0%
#74
Gemini 2.0 Flash
5.0%
#75
Cohere Command R+
5.0%
#76
Claude 3.7 Sonnet
5.0%
#77
Claude 3.5 Sonnet
4.0%
#78
Claude 3.5 Haiku
4.0%
#79
Claude 4.0 Sonnet
4.0%
#80
DeepSeek V3
4.0%
#81
Mistral Medium 3.1
4.0%
#82
Mistral Large 3
4.0%
#83
Devstral 2
4.0%
#84
GPT-4o Mini
4.0%
#85
Claude Haiku 4.5
4.0%
#86
Claude 3 Sonnet
4.0%
#87
GPT-5 Nano
4.0%
#88
GPT-5.4 Nano
4.0%
#89
Phi-4
4.0%
#90
Mistral Large 2
4.0%
#91
Llama 3.1 405B
4.0%
#92
Llama 3.3 70B
4.0%
#93
Devstral Small 2
3.0%
#94
Claude 3 Opus
3.0%
#95
GPT-4o
3.0%
#96
Mistral Large
3.0%