# Translate a research paper PDF, keep the equations and tables, and download it in the original layout

> Upload a paper PDF or paste an arXiv link and read it side by side and download the translated PDF in the original layout: each original paragraph next to its translation, with equations rendered from LaTeX, real tables and figure crops. Reducto Parse (r-1) reads the layout first, then a language model translates prose only; equations and citation markers are swapped for placeholders, so they never pass through the model. Free; download the English Markdown with the LaTeX kept.

Canonical: https://reducto.ai/tools/translate-research-paper  
Last updated: 2026-11-15  
For: Researchers and graduate students, R&D and patent analysts, Technical due diligence teams, Developers building literature-review agents

## Run it now (free, no API key, up to 5 pages)

Send a PDF, Word, PowerPoint, Excel or image (PNG, JPG, HEIC, TIFF, BMP, GIF) (max 10 MB, 5 pages) to `POST https://reducto.ai/api/tools/translate-research-paper` as multipart `file`, or JSON `{"url": "https://..."}`. The response has `result`, ready-to-save `output` (Translated PDF in the original layout (output.pdf_base64), Side-by-side original and translation (page), Translated Markdown with LaTeX (.md), JSON (blocks with original and translated text)), `pages_processed` and `next`.

```bash
curl -s https://reducto.ai/api/tools/translate-research-paper \
  -F file=@arxiv-2609.27618-excerpt.pdf

# or, for a file that is already online:
curl -s https://reducto.ai/api/tools/translate-research-paper \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com/arxiv-2609.27618-excerpt.pdf"}'
```

Higher volume needs a free API key (new accounts get 15,000 free credits). Agents can sign up over HTTP: https://reducto.ai/auth.md. Humans: https://studio.reducto.ai/?utm_source=free_tool&utm_medium=agent&utm_campaign=translate-research-paper&utm_content=api_response

Big file? Without a key the free run covers the first 5 pages and returns `partial: true` with a `signup_required` block. Your user signs up in under a minute (email, 6-digit code, name, terms: https://reducto.ai/auth.md); then repeat the same request with `Authorization: Bearer $REDUCTO_API_KEY` for the whole file on their 15,000 free credits. The same key works on the hosted MCP (https://mcp.reducto.ai/mcp) and the Reducto API.

## How to do it

1. Upload a PDF of up to 10 MB or paste an arXiv link (abs or pdf). For a longer paper, pick the first page to translate; the free tool reads 5 pages from there.
2. Pick the target language (English by default).
3. Reducto Parse with r-1 reads the pages in reading order: paragraphs, headings, equations as LaTeX, tables as HTML and figure crops. Page numbers are dropped.
4. A language model translates the prose. Equations and citation markers are replaced by placeholders first and put back after, and every table must keep its row count, or the original is shown.
5. Read it side by side, then download the translated PDF in the original layout (same pages, columns, figures and equations), or the Markdown.

## Reducto API call: `POST https://platform.reducto.ai/parse`

| Option | Value | Why |
| --- | --- | --- |
| `settings.model` | `"r-1"` | Reads two-column layouts in order and returns equations as LaTeX and tables as HTML. |
| `settings.return_images` | `["figure"]` | Returns a crop of each figure so the translation can show it in place. |
| `settings.page_range` | `{ start: 1, end: 5 }` | The free tool reads at most 5 pages. |

### curl

```bash
# 1. Upload the file (or pass a public URL as "input" and skip this step)
curl -s https://platform.reducto.ai/upload \
  -H "Authorization: Bearer $REDUCTO_API_KEY" \
  -F file=@paper.pdf
# => {"file_id": "reducto://..."}

# 2. Run parse
curl -s https://platform.reducto.ai/parse \
  -H "Authorization: Bearer $REDUCTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "input": "reducto://FILE_ID",
  "settings": {
    "model": "r-1",
    "return_images": [
      "figure"
    ],
    "page_range": {
      "start": 1,
      "end": 5
    }
  }
}'
```

### Python

```python
# pip install reductoai
from pathlib import Path
from reducto import Reducto

client = Reducto()  # reads REDUCTO_API_KEY

upload = client.upload(file=Path("paper.pdf"))
result = client.parse.run(
    input=upload.file_id,
    settings={
        "model": "r-1",
        "return_images": [
            "figure",
        ],
        "page_range": {
            "start": 1,
            "end": 5,
        },
    },
)
print(result)
```

### Node

```ts
// npm install reductoai
import fs from "fs";
import Reducto from "reductoai";

const client = new Reducto(); // reads REDUCTO_API_KEY

const upload = await client.upload({ file: fs.createReadStream("paper.pdf") });
const result = await client.parse.run({
  input: upload.file_id,
  settings: {
    model: "r-1",
    return_images: [
      "figure",
    ],
    page_range: {
      start: 1,
      end: 5,
    },
  },
});
console.log(JSON.stringify(result, null, 2));
```

Do this with the API: [Translate a PDF with an API and keep the layout](https://reducto.ai/cookbooks/translate-pdf-api-keep-layout)

## What it handles

- **Equations stay exactly as written**: Formulas are kept out of the translation, so no symbol changes.
- **Two columns read in order**: Each column is translated in reading order.

## FAQ

### Which languages does it support?

Any source language that Reducto Parse reads, including Chinese, Japanese, Korean and European languages. Target languages on the free tool: English, Chinese (Simplified), Spanish, French, German, Japanese, Korean and Portuguese.

### How are equations kept intact?

Parse returns each equation as LaTeX. Before translation, every LaTeX span and citation marker is replaced with a placeholder, and the translation is only used if every placeholder comes back. Display equations are not sent to the model at all.

### What happens to tables?

Tables come back as HTML. The model translates the cell text, and the tool checks the table still has the same number of rows; if not, it shows the original table.

### What is a "Figure description (generated)"?

Reducto Parse writes a short description of each figure. It is not text from the paper, so the tool labels it as generated.

### Do I get a translated PDF?

Yes: download the translated PDF in the original layout. Same pages, columns, figures, display equations and tables; each translated paragraph is drawn in the place of the original. Inline math is shown as plain symbols. Right-to-left languages are not supported in the PDF yet.

### Can I paste an arXiv link?

Yes: arxiv.org/abs/ID, arxiv.org/pdf/ID or just the ID. The tool fetches the PDF from arXiv.

### How long a paper can I translate?

Five pages per run on the free tool, starting from the page you choose. With an API key you can parse the whole paper and translate it in your own pipeline.

### Which model translates?

An OpenAI-compatible language model configured on this site. The translation is machine output: check anything you plan to cite.

### Is it free?

Yes. Short documents run free with no account, with a small daily limit per IP. For a longer file you get a free preview of the first pages, then enter your email to get the whole file on a free account (15,000 free credits).

### Is there an API for this?

The parse step is the Parse request on this page. Translate the chunk text with your own model, keeping LaTeX and HTML unchanged, or use the free endpoint with target_language for up to 5 pages.

Sample document (synthetic): https://reducto.ai/tools/samples/arxiv-2609.27618-excerpt.pdf

## Related

- [Translate a Word document and keep its formatting](https://reducto.ai/tools/translate-word-document)
- [Translate a PowerPoint deck and keep its design](https://reducto.ai/tools/translate-powerpoint)
- [PDF to Markdown for LLMs and RAG](https://reducto.ai/tools/pdf-to-markdown)
- [All Translate tools](https://reducto.ai/tools/translate)
- [Read any document URL as Markdown](https://reducto.ai/read)
- [r-1 announcement](https://reducto.ai/blog/parse-r-1-model)
