TOON vs JSON — Complete Comparison with Token Counts

Summary: when to use each format

Example 1 — User database (high savings)

This is the most common real-world case: an array of objects all sharing the same keys. Token counts use the GPT o200k_base tokenizer.

JSON — ~340 tokens

{
  "users": [
    {"id": 1, "name": "Alice Chen", "email": "alice@example.com", "role": "admin", "active": true},
    {"id": 2, "name": "Bob Smith", "email": "bob@example.com", "role": "user", "active": true},
    {"id": 3, "name": "Carol White", "email": "carol@example.com", "role": "user", "active": false},
    {"id": 4, "name": "Dan Brown", "email": "dan@example.com", "role": "editor", "active": true},
    {"id": 5, "name": "Eve Davis", "email": "eve@example.com", "role": "user", "active": true}
  ]
}

TOON — ~155 tokens (54% fewer)

users[5]{id,name,email,role,active}:
  1,Alice Chen,alice@example.com,admin,true
  2,Bob Smith,bob@example.com,user,true
  3,Carol White,carol@example.com,user,false
  4,Dan Brown,dan@example.com,editor,true
  5,Eve Davis,eve@example.com,user,true

The saving comes from writing the five field names once instead of five times. With 50 or 500 rows, the saving per-row stays constant — so larger arrays benefit even more.

Example 2 — Simple config object (moderate savings)

Config objects without arrays save less, but still benefit from the removal of quoted keys and surrounding braces.

JSON — ~65 tokens

{
  "model": "gpt-4o",
  "temperature": 0.7,
  "max_tokens": 1024,
  "stream": true,
  "system_prompt": "You are a helpful assistant."
}

TOON — ~48 tokens (26% fewer)

model: gpt-4o
temperature: 0.7
max_tokens: 1024
stream: true
system_prompt: You are a helpful assistant.

Example 3 — Nested structure

Deeply nested structures benefit from TOON's indented object format, which removes quotes around keys while preserving hierarchy.

JSON — ~90 tokens

{
  "request": {
    "user": {"id": 42, "tier": "pro"},
    "query": "summarise report",
    "context_length": 8192
  }
}

TOON — ~65 tokens (28% fewer)

request:
  user:
    id: 42
    tier: pro
  query: summarise report
  context_length: 8192

Why JSON remains the right choice in many contexts

TOON is designed for LLM prompts, not as a universal JSON replacement. Keep using JSON when:

• The consumer is a standard web API, database, or config parser expecting JSON
• You want the LLM to produce structured output you will parse programmatically
• The data is a single object with fewer than five fields
• The tooling you use (LangChain, LlamaIndex, etc.) handles data transformation for you

Verify savings for your own data

Theoretical averages do not tell you what your specific data will save. The JSONtoTOON converter shows you exact token counts using the GPT o200k_base tokenizer the moment you paste your JSON. No sign-up, no server calls.