Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

CaseLaw

A specialized legal analysis benchmark developed with Jurisage that evaluates models' ability to analyze and reason about recent case law in family and criminal domains. The dataset uses private, post-training-cutoff cases from June 2024, testing models' capabilities in handling novel legal scenarios across US and Canadian jurisdictions. This benchmark is particularly valuable for assessing how models handle recent precedents and international legal systems, addressing a gap in legal LLM evaluation which has historically focused primarily on US law. The evaluation includes analysis of case patterns, precedent application, and legal reasoning across multiple jurisdictions.
Source:

Model Performance

#1
Grok-3-Beta
88.1%
#2
Grok-3-Think
88.1%
#3
DeepSeek-V3-0324
86.2%
#4
Gemini-2.5-Pro-0325
86.1%
#5
Gemini-2.5-Pro-Thinking
86.1%
#6
Claude-4.0-Sonnet-Thinking
85.3%
#7
Claude-4.0-Sonnet
85.2%
#8
Claude-3.5-Sonnet-1022
84.9%
#9
Mistral-Medium-3.1
84.9%
#10
Claude-3.7-Sonnet-Thinking
84.8%
#11
DeepSeek-V3
84.5%
#12
Qwen-3-Thinking
84.0%
#13
Qwen-3
84.0%
#14
Claude-3.5-Haiku
83.8%
#15
Grok-3-Mini-Beta
83.6%
#16
Gemini-2.5-Flash-Thinking
82.5%
#17
Claude-4.1-Opus-Thinking
82.4%
#18
GPT-5-mini
82.1%
#19
OpenAI-O1-1217
81.9%
#20
Gemini-2.5-Flash
81.5%
#21
Claude-4.0-Opus
81.5%
#22
Claude-3-Opus
81.3%
#23
OpenAI-O4-mini-high
81.1%
#24
OpenAI-O4-mini-medium
81.1%
#25
DeepSeek-R1
81.0%
#26
Claude-4.1-Opus
80.9%
#27
Claude-3.7-Sonnet
80.7%
#28
Grok-4
80.6%
#29
Grok-4-Thinking
80.6%
#30
Cohere-Command-A
79.7%
#31
OpenAI-O3-mini-medium
78.5%
#32
Claude-3-Sonnet
78.5%
#33
OpenAI-O3-mini-high
78.5%
#34
GPT-4.1
78.1%
#35
OpenAI-O1-mini
77.8%
#36
Llama-3.3-70B
76.6%
#37
Llama-4-Maverick-17B
76.1%
#38
Mistral-Large-2
75.1%
#39
Gemini-2.0-Flash-Thinking-0121
74.2%
#40
Gemini-2.0-Flash
73.7%
#41
Kimi-K2-Instruct
72.0%
#42
GPT-4-mini
70.8%
#43
OpenAI-O3-medium
69.5%
#44
OpenAI-O3-high
69.5%
#45
Gemini-1.5-Pro
67.6%
#46
Cohere-Command-R-Plus
66.9%
#47
Qwen-3-Max-Preview
65.3%
#48
GPT-3.5-Turbo
58.4%