Quick Verdict
PDFMathTranslate translates scientific and technical PDFs while trying to preserve equations, figures, annotations, contents, and multi-column layouts. The official PDFMathTranslate organization now has two lines that should not be conflated: stable pdf2zh 1.9.11 for established 1.x workflows, and BabelDOC-based pdf2zh-next 2.9.0 as the evolving 2.x line for harder layout cases. Both are AGPL-3.0. The software can be obtained at no charge, but translation providers, hosted services, and infrastructure may cost money and may receive document content.
Best For
It is best for researchers, engineers, and students who need bilingual papers rather than plain extracted text. Teams can also deploy it locally and choose a local or approved cloud translation backend. It is less suitable for a few paragraphs of ordinary text, scanned handwriting without an OCR stage, or a zero-review publishing workflow. Organizations embedding it into a closed product or hosted service should review the AGPL obligations before deployment.
Key Features
- Preserves scientific-document structure, including formulas, figures, tables of contents, and annotations where the source PDF permits.
- Produces translated-only and bilingual comparison PDFs for easier terminology review.
- Supports CLI, WebUI, and Docker workflows, with page ranges and language/provider selection available in established flows.
- Works with multiple translation backends rather than requiring one model vendor.
- Uses BabelDOC in
pdf2zh-nextto improve difficult layouts, cross-column context, compatibility, and scaling behavior. - Supports local processing paths, although privacy still depends on the selected translation backend.
Use Cases
Use it to create a Chinese-English reading copy of a research paper while retaining equation numbers and figure placement, translate selected pages of a technical report, or run an internal WebUI with an approved local model. Before batch processing a conference archive, test several representative PDFs containing two-column layouts, long equations, unusual fonts, and dense tables. Layout preservation is a best-effort conversion, not a guarantee of pixel-identical output.
Pricing
| Version or route | Software price | Main distinction |
|---|---|---|
pdf2zh 1.9.11 | Free | Stable 1.x package and established CLI behavior |
pdf2zh-next 2.9.0 | Free | Official BabelDOC-based 2.x line for evolving layout handling |
| Translation API | Provider billing | Token, character, or page charges depend on the selected service |
| Third-party hosted service | Free quota or paid | Convenience, retention, privacy, and SLA are controlled by its operator |
AGPL-3.0 is not a permissive license. Distribution and modified network-service scenarios can require corresponding source availability. The software license does not cover model charges and grants no rights to source papers, illustrations, fonts, or translated publications.
Pros
- Purpose-built for scientific PDFs rather than generic text extraction.
- Clear stable 1.x and evolving 2.x choices.
- Multiple translation providers support cost, quality, and privacy tradeoffs.
- CLI, WebUI, and Docker cover personal, batch, and team workflows.
- Bilingual output makes terminology and omission review practical.
Cons
- Complex tables, scans, unusual fonts, and malformed PDFs can still require manual correction.
- AGPL-3.0 needs careful review for closed products and hosted modifications.
- High-quality commercial translation backends can make long documents expensive.
- Cloud APIs and hosted services expand the document-data exposure surface.
- Users remain responsible for source-document copyright, provider terms, and rights to publish translated output.
Alternatives
| Alternative | Better when | Tradeoff |
|---|---|---|
| Ollama | You prioritize running a model locally | Useful as a local model layer, but not a PDF reconstruction pipeline |
| Quark | You want an accessible browser, scan, and PDF suite | Easier general workflow, with different control over scientific translation and self-hosting |
| Kimi | You want Chinese long-document chat and summaries | Better interactive analysis, not a full bilingual layout-reconstruction export |
| NotebookLM | You want source-grounded research across selected material | Stronger synthesis and source navigation, without equivalent PDF reconstruction |
| Paperpal | Academic English editing and submission checks are central | Focuses on writing quality rather than full-paper bilingual layout translation |
FAQ
Should I use 1.9.11 or 2.9.0?
Choose pdf2zh 1.9.11 for established scripts and stable 1.x behavior. Evaluate pdf2zh-next 2.9.0 for its BabelDOC architecture and harder layouts. Compare them on your own PDFs.
Is it completely free?
The code is free to obtain. Translation APIs, hosted providers, compute, and storage may be paid.
Can a company use AGPL-3.0 software?
Potentially, but distribution, modifications, and network access can create source-code obligations. Legal review should match the actual architecture.
Does local installation keep every document private?
Not if a cloud translation backend is selected. Text can still be sent to that provider, while a hosted WebUI may receive the entire file.
Can I republish every translated PDF?
No. The tool does not grant translation or republication rights for the source paper, figures, fonts, or other protected material.
Bottom Line
PDFMathTranslate is a strong choice when layout-aware bilingual output matters more than one-click simplicity. Select the correct product line, validate representative files, and document your translation provider, retention policy, AGPL position, and source rights before production use. That governance is what turns a useful translator into a dependable document workflow.