Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

AA-LCR

Artificial Analysis Long Context Reasoning (AA-LCR) Dataset includes 100 hard text-based questions that require reasoning across multiple real-world documents, with each document set averaging ~100k input tokens. Questions are designed such that answers cannot be directly retrieved from documents and must instead be reasoned from multiple information sources.
Source:

Model Performance

#1
GPT-5
76.0%
#2
GPT-5-Thinking
76.0%
#3
OpenAI-O3-medium
69.0%
#4
OpenAI-O3-high
69.0%
#5
Grok-4-Thinking
68.0%
#6
Grok-4
68.0%
#7
GPT-5-mini-Thinking
68.0%
#8
Qwen-3-Thinking
67.0%
#9
Gemini-2.5-Pro-0325
66.0%
#10
Gemini-2.5-Pro-Thinking
66.0%
#11
GPT-5-mini
66.0%
#12
Claude-4.1-Opus-Thinking
66.0%
#13
Claude-4.0-Sonnet-Thinking
65.0%
#14
Gemini-2.5-Flash-Thinking
62.0%
#15
GPT-4.1
61.0%
#16
DeepSeek-R1
55.0%
#17
OpenAI-O4-mini-medium
55.0%
#18
OpenAI-O4-mini-high
55.0%
#19
Kimi-K2-Instruct
51.0%
#20
GPT-OSS-120B
51.0%
#21
Grok-3-Mini-Beta
50.0%
#22
Gemini-2.5-Flash
46.0%
#23
Llama-4-Maverick-17B
46.0%
#24
Claude-4.0-Sonnet
44.0%
#25
GPT-5-nano-Thinking
42.0%
#26
DeepSeek-V3-0324
41.0%
#27
Qwen-3-Max-Preview
40.0%
#28
GPT-5-nano
40.0%
#29
Claude-4.0-Opus
36.0%
#30
Claude-4.0-Opus-Thinking
34.0%
#31
Qwen-3
31.0%
#32
DeepSeek-V3
29.0%
#33
Gemini-2.0-Flash
28.0%
#34
Mistral-Medium-3.1
20.0%
#35
Cohere-Command-A
18.0%
#36
Llama-3.3-70B
15.0%
#37
Mistral-Large-2
5.0%
#38
Llama-3.1-405B
0.0%
#39
Magistral-Medium-3.1
0.0%
#40
Phi-4
0.0%