Can TOON Format Cut LLM Token Costs by Up to 60%?

When large language models (LLMs) are used in business systems, structured data must be exchanged in a machine‑readable format. JSON is the de‑facto standard, but its punctuation (brackets, quotes, commas) adds a substantial token overhead, increasing API costs for high‑frequency calls.

Example JSON and token count

{
  "users": [
    {"id":1, "name":"Alice", "role":"admin"},
    {"id":2, "name":"Bob",   "role":"user"}
  ]
}

Token count: 47

TOON format

TOON replaces repeated punctuation with a one‑time field declaration and indentation. The same data can be expressed as:

users[2]{id,name,role}:
  1,Alice,admin
  2,Bob,user

Token count: 24 (≈40‑60 % fewer tokens than JSON).

Benchmark results

Across several serialization formats, TOON achieved the highest accuracy while using the fewest tokens.

TOON           ████████████████████   26.9 │ 73.9% acc │ 2,744 tokens
JSON compact   █████████████████░░░   22.9 │ 70.7% acc │ 3,081 tokens
YAML           ██████████████░░░░░░   18.6 │ 69.0% acc │ 3,719 tokens
JSON           ███████████░░░░░░░░░   15.3 │ 69.7% acc │ 4,545 tokens
XML            ████████░░░░░░░░░░   13.0 │ 67.1% acc │ 5,167 tokens

When TOON may be unsuitable

Deeply nested or irregular structures – compact JSON often uses fewer tokens.

Semi‑uniform arrays (≈40‑60 % fit a table format); if your pipeline already depends on JSON, staying with JSON may be simpler.

Pure tabular data – CSV is smaller than TOON for flat tables.

Latency‑critical applications – some deployments process compact JSON faster despite higher token counts; benchmark first‑token‑time (TTFT) and throughput.

Java integration

Add the TOON library to your Maven project. The required coordinates are:

groupId:    com.felipestanzani
artifactId: jtoon
version:    0.1.2

Core API methods:

// Java object → TOON string
String toon = JToon.encode(object);

// JSON string → TOON
String toon = JToon.encodeJson(jsonString);

// TOON → Java object
Object obj = JToon.decode(toonString);

// TOON → JSON string
String json = JToon.decodeToJson(toonString);

In scenarios with large volumes of uniform object arrays, TOON can markedly reduce token consumption and improve LLM retrieval accuracy, leading to lower API expenses.