🔥 JSON vs TOON: Which Format Wins for LLMs like GPT, Gemini & Claude?
TOON (Token-Oriented Object Notation) reimagines JSON for the AI era — built for token efficiency and readability, saving 30–60% tokens for LLM workloads.
By Escose Technologies | Nov 2025 | AI/ML
Introduction
If you've ever built prompts or pipelines for LLMs, you've definitely sent a ton of JSON down the wire.
But here's the sneaky part — Every {}, [], and " in your JSON counts as a token. That means longer prompts, higher costs, and sometimes… slower responses.
Enter TOON: Token-Oriented Object Notation
TOON is a brand-new format introduced in 2025 by Johann Schopplich. It reimagines JSON for the AI era — built for token efficiency and readability.
- No curly braces or quotes — clean indentation, table-like layout.
- 30–60% fewer tokens for most flat or tabular datasets.
- Keys declared once = lightweight + human-friendly.
- Perfect for prompts, datasets, and agent pipelines.
Why It Matters
When you're running LLMs at scale, every token saved = money saved + faster inference.
Of course, JSON isn't going anywhere — it's still king for deeply nested data. But for LLM workloads? TOON is quietly taking center stage.