You are throwing half of the token into the trash.

2026-06-22 · Source: LLM on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, medium

Summary

Uploading raw files like PDFs or Word documents to Large Language Models (LLMs) significantly increases token consumption and cost, as models process unnecessary "packaging" data such as page layouts, fonts, and binary structures. This overhead can account for up to half of paid tokens. Additionally, agglutinative languages like Turkish incur a "language tax" because tokenizers, primarily trained on English, break down suffixes into separate tokens, leading to approximately 1.5 tokens per Turkish word compared to 1 token per English word. To mitigate these issues, converting documents to Markdown format before ingestion is recommended. Markdown preserves structure while stripping extraneous data, resulting in fewer tokens, improved model accuracy by reducing noise, and faster response times. Tools like Microsoft's MarkItDown, IBM's Docling, and Pandoc facilitate this conversion.

Key takeaway

For AI Engineers optimizing LLM inference costs and improving model accuracy, you should prioritize converting input documents to Markdown format. This practice significantly reduces token consumption by eliminating unnecessary formatting data, especially crucial when processing agglutinative languages like Turkish. By adopting tools such as MarkItDown or Docling for conversion, you can achieve substantial savings on API calls and enhance the model's ability to focus on core content, leading to more reliable outputs.

Key insights

Directly uploading formatted files or using agglutinative languages to LLMs drastically inflates token costs and reduces processing efficiency.

Principles

• Tokenizers are optimized for English text.
• File formatting adds significant token overhead.
• Simpler input improves LLM focus and accuracy.

Method

Convert source documents (PDF, Word, Excel) to Markdown (.md) format using tools like MarkItDown, Docling, or Pandoc, then review and clean the converted Markdown for residual noise before feeding to LLMs.

In practice

• Use MarkItDown for general file conversion.
• Employ Docling for complex, table-heavy documents.
• Clean converted Markdown to remove page numbers.

Topics

• LLM Cost Optimization
• Tokenization Efficiency
• Markdown Conversion
• Document Preprocessing
• Agglutinative Languages
• MarkItDown Library

Best for: AI Engineer, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Your LLM Has Never Read a Single Word — How Tokenization Grinds Text Into Numbers — LLMs process numeric tokens, not words or letters, leading to inherent limitations like character blindness and variable processing costs across languages.
• microsoft / markitdown — MarkItDown converts diverse document types into structured Markdown for LLM consumption and text analysis.
• Stop Wasting Tokens: A Smarter Alternative to JSON for LLM Pipelines — TOON reduces LLM token waste by compactly representing repetitive structured data, optimizing prompt efficiency.
• If an LLM could genuinely deduce the intellectual baseline of its user from provided documents or conversational context, it could eliminate boilerplate introductions, redundant explanations, and... — LLMs' lack of true epistemic modeling drives costly verbosity and reduces accuracy, necessitating explicit mitigation strategies.
• AI 101: What Is a Token (and why it runs AI)? — Tokens are the fundamental units AI models process, impacting context, performance, and cost.
• Only about 1% of the tokens Thariq generates go into production code — AI-driven content generation reduces documentation costs, enabling focus on idea quality and flexible, just-in-time practices.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by LLM on Medium.