TOON vs. JSON Benchmarks
For a typical 100-row customer list, JSON takes approximately 12,000 tokens. Compact JSON (minified) takes 8,500 tokens. TOON takes just 4,200 tokens. That's a 65% reduction in API costs and context window usage without any loss of data clarity.
How It Works
TOON uses a header like users[100]{id,name,role}: to define the fields once. Each line that follows contains only the values. This eliminates the "key redundancy" that plagues JSON and YAML.
Frequently Asked Questions
How much money can TOON save me?
If you are processing 1 million rows of data per month through GPT-4o, switching from JSON to TOON can save you over $500/month by reducing token consumption by half.
Do LLMs find TOON hard to read?
No. In benchmarks, GPT-5.4 and Claude Sonnet 4.7 showed identical comprehension scores on TOON vs. JSON data, while using significantly fewer tokens. The format is intuitive enough for any frontier model.