TOON Reference

TOON (Token-Oriented Object Notation) is a compact, lossless re-encoding of the JSON data model designed for LLM input. It combines YAML-style indentation for nesting with CSV-style tables for uniform arrays. Spec: https://github.com/toon-format/spec. Reference implementation: https://github.com/toon-format/toon.

Why convert to TOON

• ~40% fewer tokens than pretty JSON on mixed-structure benchmarks; comparable accuracy or better.
• Lossless for the JSON data model — round-trips back to the same JSON.
• LLM-friendly guardrails — [N] array lengths and {fields} headers give the model an explicit schema.

When TOON helps

• Uniform arrays of objects (table-shaped data) — biggest savings.
• Mixed structures with some uniform arrays.
• Anything you'd otherwise drop into a prompt as JSON.

When TOON does NOT help (or hurts)