MathVision

Overview

MATH-Vision (MATH-V) is a meticulously curated dataset of 3,040 high-quality mathematical problems with visual contexts sourced from real math competitions. It evaluates mathematical reasoning abilities in multimodal settings.

Task Description

• Task Type: Visual Mathematical Reasoning

• Input: Math problem with visual context (diagram, figure, graph)

• Output: Numerical answer or answer choice

• Domains: Competition mathematics with visual elements

Key Features

• Problems sourced from real mathematical competitions

• 3,040 high-quality problems with visual contexts

• Five difficulty levels (1-5) for granular analysis

• Supports both multiple-choice and free-form answers

• Includes detailed solutions for reference

Evaluation Notes

• Default evaluation uses the test split

• Subsets by difficulty: level 1 through level 5

• Primary metric: Accuracy with numeric comparison

• Free-form answers use \boxed{} format (without units)

• Multiple-choice uses CoT prompting with letter answers

Properties

Property	Value
Benchmark Name	math_vision
Dataset ID	evalscope/MathVision
Paper	N/A
Tags	MCQ, Math, MultiModal, Reasoning
Metrics	acc
Default Shots	0-shot
Evaluation Split	test

Data Statistics

Metric	Value
Total Samples	3,040
Prompt Length (Mean)	423.91 chars
Prompt Length (Min/Max)	93 / 1581 chars

Per-Subset Statistics:

Subset	Samples	Prompt Mean	Prompt Min	Prompt Max
level 1	237	357.8	133	854
level 2	738	397.17	127	885
level 3	551	433.46	93	1402
level 4	830	434.51	134	946
level 5	684	455.12	129	1581

Image Statistics:

Metric	Value
Total Images	3,040
Images per Sample	min: 1, max: 1, mean: 1
Resolution Range	54x40 - 2520x8526
Formats	jpeg

Sample Example

Subset: level 1

{
  "input": [
    {
      "id": "606d1e5a",
      "content": [
        {
          "text": "Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of A,B,C,D,E. Think step by step before answering.\n\nWhich kite has the longest string?\n<image1>\n\nA) A\nB) B\nC) C\nD) D\nE) E"
        },
        {
          "image": "[BASE64_IMAGE: jpg, ~38.6KB]"
        }
      ]
    }
  ],
  "choices": [
    "A",
    "B",
    "C",
    "D",
    "E"
  ],
  "target": "C",
  "id": 0,
  "group_id": 0,
  "subset_key": "level 1",
  "metadata": {
    "id": "3",
    "image": "images/3.jpg",
    "solution": null,
    "level": 1,
    "question_type": "multi_choice",
    "subject": "metric geometry - length"
  }
}

Prompt Template

Prompt Template:

{question}
Please reason step by step, and put your final answer within \boxed{{}} without units.

Usage

Using CLI

evalscope eval \
    --model YOUR_MODEL \
    --api-url OPENAI_API_COMPAT_URL \
    --api-key EMPTY_TOKEN \
    --datasets math_vision \
    --limit 10  # Remove this line for formal evaluation

Using Python

from evalscope import run_task
from evalscope.config import TaskConfig

task_cfg = TaskConfig(
    model='YOUR_MODEL',
    api_url='OPENAI_API_COMPAT_URL',
    api_key='EMPTY_TOKEN',
    datasets=['math_vision'],
    dataset_args={
        'math_vision': {
            # subset_list: ['level 1', 'level 2', 'level 3']  # optional, evaluate specific subsets
        }
    },
    limit=10,  # Remove this line for formal evaluation
)

run_task(task_cfg=task_cfg)