MathVista

multimodal

MathVista tests mathematical reasoning in visual contexts using 6,141 examples from 31 datasets. It evaluates whether models can solve maths problems involving diagrams, graphs, charts, and geometric figures.

View paper / source

0

Models Tested

0.0

Average Score

0–100

Scale Range

1x

Weight

How It Works

Models receive images containing mathematical content (function plots, geometry diagrams, statistical charts) and must solve problems that require both visual understanding and mathematical reasoning.

Why It Matters

Mathematical reasoning with visual inputs is a critical skill for STEM applications. MathVista uniquely combines vision and mathematics, testing a capability that pure text benchmarks miss entirely.

Limitations

Image quality and format can affect results. Some problems may be solvable from text descriptions alone. The dataset draws from existing sources which may appear in training data.

Leaderboard — MathVista

No model scores recorded yet for this benchmark.
All Benchmarks