arxiv-md

Good enough™ LaTeX to Markdown converter. Optimized on hundreds of arXiv papers from 2016 to 2026.

arxiv-md is a Python package and CLI for turning arXiv e-print bundles, source directories, or single .tex files into:

- document.md       # rendered Markdown
- conversion.json   # warnings, stats, options, paths
- images/           # optional rendered/copied figures

Unsupported TeX is preserved as raw LaTeX when possible and reported through typed diagnostics. Output locations are explicit: callers choose the exact output directory, and conversion never writes into the source tree.

Install

# Core library and CLIs
uv add arxiv-md

# Add optional pypdfium2/Pillow-backed PDF/JPEG → PNG rendering
uv add 'arxiv-md[assets]'

# CLI-only install
uv tool install 'arxiv-md[assets]'

Supports Python 3.10–3.13. Core library has no required runtime dependencies.

Quickstart

# Single .tex file
tex-to-md paper.tex --outdir out/paper

# Source archive (.tar, .tar.gz, .tgz, .zip, .gz)
tex-to-md paper.tar.gz --outdir out/paper --json

# Download arXiv source by id and convert
arxiv-to-md 1706.03762 --outdir out

--outdir is mandatory. tex-to-md writes one document directly into that folder. arxiv-to-md writes one subdirectory per paper.

Common asset modes:

• Default rasterize — PDF→PNG via pypdfium2, JPEG→PNG via Pillow; best compatibility, highest CPU/native-code exposure.
• --asset-mode copy — copy PDF/JPEG verbatim; faster, no optional asset deps needed.
• --asset-mode skip — resolve/count figures but write no images.
• --no-assets — text-only conversion; no images/ directory.

Asset rendering extra uses pypdfium2 + Pillow; core install has no required deps and no AGPL PyMuPDF runtime dependency.

Full CLI reference: docs/cli.md.

Python API

from pathlib import Path

from arxiv_md import ConvertOptions, convert_path, write_result

result = convert_path(Path("paper.tex"), ConvertOptions(render_assets=False))
print(result.markdown)
write_result(result, "out/paper")

convert_path returns ConvertResult with rendered Markdown, typed document IR, warnings, stats, and output paths when files are written.

Full API reference: docs/api.md.

At a glance

Inputs:

• .tex files.
• Source directories with a detectable main .tex file.
• .tar, .tar.gz, .tgz, .zip, and single-file .gz source bundles.
• arXiv IDs or search queries via arxiv-to-md.

Native handling covers common paper structure: sections, frontmatter, inline formatting, math, lists, figures, tables, bibliography, citations, cross-references, theorem-like environments, algorithm pseudocode, common glyphs, siunitx, and common macro definitions.

Unsupported inputs (.pdf, .html, .rar, .7z, lone .bib).

When to use it

Use arxiv-md when you need semantic Markdown for search, indexing, previews, datasets, or downstream text processing.

Do not use it when exact PDF layout matters. arxiv-md is not a TeX engine: it does not run pdflatex, reproduce page layout, or guarantee full macro and environment expansion.

Docs

• docs/README.md — documentation index.
• docs/cli.md — CLI usage, flags, output layout.
• docs/api.md — library options, result shape, document IR.
• docs/supported-latex.md — inputs, non-goals, support matrix.
• docs/diagnostics.md — warnings, fatal errors, JSON envelopes.
• docs/security.md — archive hardening, isolation, resource limits.
• docs/performance.md — benchmarks and asset-mode tradeoffs.

License

MIT. See LICENSE.