How can an AI agent translate an arXiv PDF without breaking equations?

Use a layout-aware PDF translation pipeline rather than extracting plain text. The translate_pdf MCP tool accepts a public PDF URL and returns translated-only and bilingual PDFs while preserving equations, figures, and columns.

Agent 怎么翻译 arXiv PDF 而不破坏公式和双栏排版？

使用先解析版面再翻译的 PDF 流水线，而不是把文档压成纯文本。translate_pdf 接收公开 PDF URL，完成后返回保留公式、图片和分栏的单语及双语 PDF。

Measured evidence: pdf-translate

Updated 2026-08-01; 4 models measured, 3 passed every task; the lowest measured cost among complete passes was claude-sonnet-5 at $0.002774 per benchmark run.

Human-readable benchmark · Raw JSON data

Agent workflow

  1. Provide a direct public PDF URL.
  2. Choose the target language and page range.
  3. Call translate_pdf and keep the returned job_id.
  4. Follow check_job next_action until the bilingual download is ready.

Recommended tool: translate_pdf

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "translate_pdf",
    "arguments": {
      "url": "https://arxiv.org/pdf/1706.03762",
      "lang_to": "Simplified Chinese",
      "pages": "first5"
    }
  }
}

MCP endpoint: https://ainetcafe.com/mcp · Machine-readable guide

Method and reuse

Live application tasks are run repeatedly; pass/fail, actual billed USD cost, and latency are recorded. Upstream-unavailable models are excluded from quality conclusions.

Data license: CC BY 4.0. Cite the page URL and measurement date.

Updated 2026-08-01 · All benchmark data · Agent capability index