Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

LiveBench (Reasoning)

Evaluates performance on LiveBench reasoning tasks, including harder versions of Big-Bench Hard's Web of Lies and Zebra Puzzles.
Source:

Model Performance

#1
GPT-5
98.2%
#2
OpenAI o3 (High Effort)
94.7%
#3
Gemini 2.5 Pro
93.7%
#4
DeepSeek R1
91.1%
#5
OpenAI o3 (Medium Effort)
91.0%
#6
Claude 4.0 Opus (Thinking)
90.5%
#7
Claude Opus 4.6 (Thinking)
88.7%
#8
OpenAI o4 Mini (High Effort)
88.1%
#9
Grok 3 Mini
87.6%
#10
Claude Sonnet 4.6
84.8%
#11
Gemini 3.1 Pro Preview
84.0%
#12
GPT-5.1 Codex Max
83.7%
#13
GPT-5 Mini
82.6%
#14
GPT-5 (Thinking)
81.7%
#15
Claude Opus 4.5 (Thinking)
80.1%
#16
Gemini 2.5 Flash
78.5%
#17
OpenAI o4 Mini (Medium Effort)
78.5%
#18
Claude Sonnet 4.5 (Thinking)
77.6%
#19
DeepSeek V3.2 (Thinking)
77.2%
#20
Claude 3.7 Sonnet (Thinking)
76.2%
#21
Claude 4.1 Opus (Thinking)
72.3%
#22
Gemini 2.5 Pro (Thinking)
70.8%
#23
Claude 4.0 Sonnet (Thinking)
69.0%
#24
GPT-5 Mini (Thinking)
68.3%
#25
Claude Haiku 4.5 (Thinking)
61.7%
#26
Qwen 3 (Thinking)
59.4%
#27
DeepSeek V3
59.2%
#28
Qwen 3 Max Preview
58.4%
#29
Phi-4
57.8%
#30
Claude 4.0 Opus
56.4%
#31
GPT-5 Nano
54.9%
#32
GPT-4.5
54.4%
#33
Claude Opus 4.5
53.2%
#34
Gemini 2.5 Flash (Thinking)
51.5%
#35
Claude 3.7 Sonnet
49.1%
#36
GPT-4o
48.8%
#37
Qwen 3
48.2%
#38
GPT-4.1
44.4%
#39
DeepSeek V3.2
44.2%
#40
Llama 4 Maverick 17B
43.8%
#41
Claude 3.5 Sonnet
43.2%
#42
GPT-5.2
42.8%
#43
Claude Sonnet 4.5
42.3%
#44
Kimi K2
42.2%
#45
Mistral Medium 3.1
42.0%
#46
Claude 4.1 Opus
40.9%
#47
GPT-5 Nano (Thinking)
40.3%
#48
Claude 4.0 Sonnet
39.7%
#49
GPT OSS 120B
39.2%
#50
Cohere Command A
36.3%
#51
Claude Haiku 4.5
33.9%
#52
Mistral Large
33.8%
#53
Claude 3.5 Haiku
26.2%
#54
Cohere Command R+
21.6%