---
pretty_name: HistGramBench
license: apache-2.0
annotations_creators:
- machine-generated
# languages inferred from work titles (column language_title_heuristic); most rights records say `und`
language:
- en
- fr
- de
- ru
- zh
- it
- pt
- gu
task_categories:
- image-to-text
tags:
- benchmark
- tikz
- mathematics
- geometry
- historical-documents
- digital-humanities
- vision-language-models
size_categories:
- 1K<n<10K
configs:
- config_name: items
  default: true
  data_files:
  - split: test
    path: data/items/test-*.parquet
  - split: development
    path: data/items/development-*.parquet
- config_name: answer_key_facts
  data_files:
  - split: test
    path: data/answer_key_facts/test-*.parquet
  - split: development
    path: data/answer_key_facts/development-*.parquet
- config_name: answer_key_entities
  data_files:
  - split: test
    path: data/answer_key_entities/test-*.parquet
  - split: development
    path: data/answer_key_entities/development-*.parquet
- config_name: answer_key_records
  data_files:
  - split: test
    path: data/answer_key_records/test-*.parquet
  - split: development
    path: data/answer_key_records/development-*.parquet
- config_name: answers
  data_files:
  - split: test
    path: data/answers/test-*.parquet
  - split: development
    path: data/answers/development-*.parquet
- config_name: verdicts
  data_files:
  - split: test
    path: data/verdicts/test-*.parquet
  - split: development
    path: data/verdicts/development-*.parquet
- config_name: cgx
  data_files:
  - split: test
    path: data/cgx/test-*.parquet
  - split: development
    path: data/cgx/development-*.parquet
- config_name: compiled_pdfs
  data_files:
  - split: test
    path: data/compiled_pdfs/test-*.parquet
  - split: development
    path: data/compiled_pdfs/development-*.parquet
- config_name: scores
  data_files:
  - split: test
    path: data/scores/test-*.parquet
  - split: development
    path: data/scores/development-*.parquet
- config_name: works
  data_files:
  - split: records
    path: data/works/records-*.parquet
- config_name: split_groups
  data_files:
  - split: records
    path: data/split_groups/records-*.parquet
---

# HistGramBench

[![arXiv](https://img.shields.io/badge/arXiv-2611.00081-black?style=flat-square&color=black&logo=arxiv&logoColor=white)](https://arxiv.org/abs/2611.00081) [![Blog](https://img.shields.io/badge/Blog-vietalpha.org-black?style=flat-square&color=black)](https://www.vietalpha.org/research/histgrambench) [![Article](https://img.shields.io/badge/Article-Deerfield%20Focal%20Point-black?style=flat-square&color=black)](https://www.deerfieldfocalpoint.com/articles/histgrambench-can-ai-pass-mr-dollhopfs-geometry-class) [![Code](https://img.shields.io/badge/Code-GitHub-black?style=flat-square&color=black&logo=github&logoColor=white)](https://github.com/VietAlpha/histgrambench) [![License](https://img.shields.io/badge/License-Apache%202.0-black?style=flat-square&color=black)](LICENSE)

HistGramBench measures how well vision-language models redraw printed historical mathematical diagrams as TikZ.
Each figure comes with an answer key of typed facts: points, lines, circles, incidences, parallels, labels. A model's
TikZ is compiled in a sandbox, its drawing is read back as vector geometry, and each key fact is checked on that
geometry or, where geometry cannot decide, by judge models. A figure's score is the share of its key facts the
drawing recovers.

Code, protocol and reproduction instructions: [github.com/VietAlpha/histgrambench](https://github.com/VietAlpha/histgrambench).

**Paper.** [arXiv:2611.00081](https://arxiv.org/abs/2611.00081). Overview: [vietalpha.org/research/histgrambench](https://www.vietalpha.org/research/histgrambench).

## At a glance

| | |
|---|---|
| Scored figures | 2,507 from 95 works, printed 1551–1933 |
| Development figures | 150 from 5 works |
| Answer-key facts | 114,403 on the scored figures |
| Contestant answers | 30,084 (6 models × 2 tracks × 2,507 figures) |
| Tracks | A: image only; B: image plus page text |

## Loading

```python
from datasets import load_dataset

items = load_dataset("VietAlphaLabs/HistGramBench", "items", split="test")
keys = load_dataset("VietAlphaLabs/HistGramBench", "answer_key_facts", split="test")

row = items[0]
row["image"]                      # the figure given to the model
facts = keys.filter(lambda f: f["item_id"] == row["item_id"] and f["in_key"])
```

## Contents

| Config | Content |
|---|---|
| `items` | figures with image, page text and metadata |
| `answer_key_facts`, `answer_key_entities`, `answer_key_records` | answer keys and how they were built |
| `answers` | contestant outputs with outcome codes |
| `compiled_pdfs`, `cgx` | compiled drawings and the geometry read from them |
| `verdicts`, `scores` | per-fact verdicts and per-figure scores |
| `works`, `split_groups` | source works and the source groups behind the split |

Splits: `test` is the scored set; `development` was used to fix thresholds before the test run. `item_id` joins all
tables. Column definitions are in [SCHEMA.md](SCHEMA.md).

## Licence and citing figures

Apache-2.0 ([LICENSE](LICENSE)). Most figures come from works held by Wikimedia Commons, the Internet Archive, the
Göttingen Digitization Center, Numdam and Project Gutenberg. When you reproduce a figure, credit its source work
and repository as listed in `works`.

## Citation

```bibtex
@misc{tran2026histgrambench,
  title  = {{HistGramBench}: Measuring Structural Recovery When Multimodal Models Redraw Historical Mathematical Diagrams as {TikZ}},
  author = {Tran, Minh Quang and He, Carter and Bloch, Cameron and Huang, Zitai and Vo, Thien Truong and Tran, Trong Nguyen and Chen, Isaac},
  year   = {2026},
  eprint = {2611.00081},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url    = {https://arxiv.org/abs/2611.00081}
}
```