Token costs

FileConcat has output controls that change how much of your model's context window the output consumes. Some moves cost tokens rather than save them. This page tells you which is which, with measured numbers where they exist.

Show line numbers

FileConcat leaves line numbers off, and the current UI does not surface a toggle for them. The engine can still prefix every line of every file with its 1-based line number; that transform lives in packages/core/src/file-processing/transform.ts as addLineNumbers.

Measured against 8 representative repo files on 2026-06-07, turning it on raised the token count by 44.83 percent on average, roughly 40 to 50 percent per file. That cost is why it stays off by default: line numbers are a navigation aid for when you want the model to refer back to a specific line, not a saving.

Output style

The result screen's Format control offers XML, Markdown, and Plain. The XML style wraps each file in named tags and tends to read better for Claude. The Markdown style uses fenced code blocks and tends to read better for chat-style assistants that emit Markdown themselves. Plain concatenates the files with lightweight separators and no markup.

The choice costs almost nothing. Measured across five real codebases on 2026-08-31, the spread between the three styles was 0.24% to 1.13% of the bundle, and the wrapper itself ran 0.6% to 3.8%. The 60 repository measurement puts the wrapper at a median 2.4%, reaching 17.8% on a repository of many small files, where there are more file tags than file contents.

So pick on fidelity and readability, not on cost. XML vs Markdown has the per-format numbers and the published research on whether the wrapper helps a model at all. The short version: what those numbers do not say is which style a model answers better from. That has been measured for tables and for nested data, with results that contradict each other, and never for a bundle of source files, by us or by anyone else.

Cross-link

For the tokenizer itself, the sampling that takes over above 1 MB, and the model selector, see Token estimation.

For what a full context window costs to fill, priced across the whole model catalogue, see Context window costs.