◆Scientific LLM Benchmarks
GitHub
← All benchmarks
Math· multimodal-competition-math

MATH-Vision (MATH-V)

CUHK MMLab / Shanghai AI Lab · 2024

3,040 real competition problems with visual context, spanning 16 mathematical disciplines and five difficulty levels.

Mathematics
GitHub stars
Task type
VQA
Modality
multimodal
Access
open
Size
3,040 items
License
MIT
Metrics
accuracy
id
1
question
Which number should be written in place of the question mark? <image1>
options
image
images/1.jpg
decoded_image
<image/binary>
answer
60
level
2
subject
arithmetic
id
2
question
Which bike is most expensive? <image1>
options
A B C D E
image
images/2.jpg
decoded_image
<image/binary>
answer
A
level
2
subject
arithmetic

Real rows from the Hugging Face datasets server · long values truncated