PDF to Markdown for ChatGPT & Claude: the measured token savings

7 August 2026 · Milldown team · All posts

Every converter site repeats the same claim: “Markdown uses 30–60% fewer tokens.” Few say where the number comes from, and none break it down by format. We did the measurement — with the o200k tokenizer, on real documents — and the truth is more interesting: some formats save far more than 60%, and one format saves nothing at all.

The numbers, per format

Methodology: every row counts both sides on real files. The PDF baseline is the input-token count Anthropic’s own token-counting endpoint reports for a native upload (claude-opus-5, 13 August 2026), not an estimate of it. Token counts use OpenAI’s o200k tokenizer as a consistent cross-provider estimate — Claude tokenizes differently, and ratios are far more stable than absolute counts. The corpus, baselines, and per-file results are published on our benchmarks page. Milldown shows a live token count for every file you convert. The app estimates the native-PDF baseline from a per-page constant rather than calling the provider, so its figure is close to the table above without being identical, and it is deliberately the more conservative of the two.

Why PDF is the headline number

Uploading a PDF natively bills every page as extracted text plus a page image — 1,694–1,878 tokens per page, measured with Anthropic’s own token counter. Once the text is out, that per-page image cost is pure overhead, and it’s re-billed on every chat submission. Sparse pages save the most, because the fixed image cost dominates. Details in our PDF to Markdown guide.

Why we claim nothing for Word and PowerPoint

You’ll see other tools claim ~97% savings for Office files by comparing Markdown against the raw OOXML inside the zip. Our benchmark reproduces that number — 94.1% and 96.8% on real documents — and we still don’t claim it, because no chat workflow pays that baseline: every tool extracts the text before the model sees it, and against that, Markdown costs about the same. What conversion buys you for Office files is structure — real headings and tables instead of a flattened wall of text — plus cleanup and measurement. If your RAG loader genuinely ingests raw file contents, the arithmetic is on the benchmarks page; decide for yourself whether that baseline is yours. More in the Word guide.

The exception: spreadsheets

Markdown tables are more expensive than CSV — all those aligned pipes and dashes add up. On our benchmark sheets, CSV comes out 12–29% smaller than the equivalent Markdown table. That’s exactly why Milldown has a dedicated CSV output mode for Excel files: the right target format depends on the source. Use Markdown tables when a human will read them; use CSV when a model will.

Try it on your own documents

Drop your own files into Milldown and the token badge shows the before/after for each one — computed on your computer, with nothing uploaded.