Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

MortgageTax

A multimodal benchmark that evaluates language models' ability to extract and process information from mortgage tax certificates. The benchmark tests both semantic extraction (year, parcel number, county) and numerical extraction (calculating annualized amounts due) based on tax certificate images.
Source:

Model Performance

#1
Claude-3.7-Sonnet
80.6%
#2
Claude-3.7-Sonnet-Thinking
79.2%
#3
GPT-4.1
79.0%
#4
Gemini-2.5-Pro-Thinking
78.8%
#5
Gemini-2.5-Pro-0325
78.8%
#6
OpenAI-O3-medium
78.4%
#7
OpenAI-O3-high
78.4%
#8
Claude-3.5-Sonnet-1022
78.1%
#9
OpenAI-O4-mini-medium
77.1%
#10
OpenAI-O4-mini-high
77.1%
#11
GPT-5-mini
75.4%
#12
Claude-4.0-Sonnet
74.6%
#13
Llama-4-Maverick-17B
72.7%
#14
Gemini-2.0-Flash
72.6%
#15
Gemini-1.5-Pro
71.0%
#16
Gemini-2.5-Flash
70.9%
#17
Gemini-2.5-Flash-Thinking
69.5%
#18
GPT-4-mini
69.2%
#19
Claude-4.1-Opus
65.2%
#20
Claude-4.0-Sonnet-Thinking
62.5%
#21
Mistral-Medium-3.1
60.5%
#22
Claude-4.0-Opus
59.0%
#23
Grok-4
57.5%
#24
Grok-4-Thinking
57.5%
#25
Claude-4.1-Opus-Thinking
53.4%