Toon

TOON, short for Token Oriented Object Notation, is a compact, human readable way of encoding JSON data for use in large language model prompts. It represents the same objects, arrays and primitives as JSON with deterministic, lossless round trips, but uses indentation instead of braces and minimizes quoting, giving a layout closer to YAML in readability and CSV in compactness. It is aimed at developers who send structured data to models and want to spend fewer tokens. For arrays of uniform objects, TOON declares the field list once and then lists one row per item. Length markers in square brackets and field lists in braces act as guardrails, so truncated or malformed model output is easier to catch. The site reports that on its benchmarks TOON matched JSON's retrieval accuracy, 72.2 percent against 71.4 percent, while using 42.6 percent fewer tokens. The project provides a specification, official implementations and dozens of community ports in many languages that share one conformance test suite. The site also offers a playground, a command line tool, benchmark results, notes on what changed in version 4, and documentation bundles formatted for language models. The site publishes no pricing and lists no paid plans.