The fastest PDF to Markdown API — benchmarks vs PyMuPDF4LLM, Marker, and docling

TL;DR: AnyMD converts PDFs to Markdown in 0.3–0.8 seconds for typical documents — 4–10× faster than alternatives like PyMuPDF4LLM, Marker, and docling. No queuing, no async jobs, no waiting. Sync API, in-memory processing, under a second.

We built AnyMD in Rust for a reason. Every other document-to-markdown tool on the market either runs as a local Python library (blocking, single-threaded) or wraps a queue worker (latency from polling or async job dispatch). AnyMD is a compiled Rust binary with hand-tuned document parsing — your file goes in, Markdown comes back in the same HTTP response, usually in under a second.

Benchmarks

We tested conversion time for three common document types across four tools. All tests ran on the same hardware (8-core Xeon, 16 GB RAM, SSD). Times are in seconds, lower is better.

PyMuPDF4LLM and Marker only support PDF input. docling supports PDF and Office formats but is Python-based with significant overhead per document. AnyMD handles all formats natively and at near-constant latency regardless of input type.

Why Rust matters for document conversion

Document parsing is CPU-bound work — extracting text, parsing XML trees inside DOCX/PPTX files, reconstructing layout from PDF operators. Python libraries pay a GIL penalty for every operation. A Rust binary compiles to native code with zero overhead, zero garbage collection pauses, and full thread utilisation.

The difference isn't academic. A 48-page business report converts in 0.8 seconds with AnyMD vs 6–14 seconds with Python-based tools. In a RAG pipeline processing thousands of documents, that difference compounds massively.

Try it

curl https://anymd.net/api/convert \
  -H "Authorization: Bearer your-key" \
  -F "file=@report.pdf"

Free tier: 100 pages/month, no credit card required. Get your API key — it takes about 30 seconds.