PDFMathTranslate (pdf2zh) is an open-source tool with 37,000+ stars on GitHub that translates scientific PDF documents while preserving formulas, charts and the original layout.
Running a research paper through a generic translator usually destroys it: equations scramble, tables collapse and figure captions detach. PDFMathTranslate is built specifically for scientific documents — it translates the full text while keeping formulas, charts, tables of contents and annotations in place, producing clean monolingual or side-by-side bilingual PDFs. Translation itself is pluggable: connect Google, DeepL, OpenAI, Ollama and many other services, in many source and target languages. You can drive it from the command line, an interactive GUI, Docker, an MCP server, or a Zotero integration for one-click paper translation inside a reference manager. The project was presented as an EMNLP 2025 demo and is under intense active development — 2026 updates added experimental OCR support and an ultrafast mode for native-text PDFs. A free online version at pdf2zh.com handles small files in the browser.
PDFMathTranslate is free and open source under the AGPL-3.0 license, and the hosted web demo is free for small files. If you connect a commercial translation API, you pay that provider’s normal usage rates; local backends such as Ollama are free.
Researchers, graduate students and engineers who read papers across languages and need translations that keep the mathematics and figures exactly where they belong.