Input JSON(auto)
or drop a file anywhere
Output TOON
Your TOON appears here as you type.
users[3]{id,name,role}:
1,Alice,admin
2,Bob,user
3,Cleo,userPaste data to see token savings
TOON (Token-Oriented Object Notation) is a compact, human-readable encoding of the JSON data model: it declares field names once and writes uniform arrays as rows, so the same data costs fewer tokens in an LLM prompt. json2toon.co turns JSON, CSV, YAML, XML and TOML into TOON or TONL, and back again. The data model stays the same; only the repeated syntax goes.
TOON uses about 43% fewer tokens than JSON on average (42.6% in the official TOON v4.1 benchmark), which lowers API costs for ChatGPT, Claude, Gemini, and similar models. It stays readable, unlike minified JSON. The converter shows the real token count for your own data, so you can check before you switch. Read the cost optimization guide or compare JSON vs TOON and TOON vs TONL.
Everything runs client-side: nothing you paste is uploaded, and share links keep the data in the URL fragment, which browsers never send to a server. Convert many files at once with batch mode (ZIP export), automate it with the free TOON API or the TypeScript library, or convert selected text on any page with the TOON Chrome extension. Common uses: LLM optimization, chatbot and RAG pipelines, data analysis.
[4]{id,name,email,role,active}:
1,Alice,alice@example.com,admin,true
2,Bob,bob@example.com,user,true
3,Cleo,cleo@example.com,user,false
4,Dev,dev@example.com,editor,true[
{
"id": 1,
"name": "Alice",
"email": "alice@example.com",
"role": "admin",
"active": true
},
{
"id": 2,
"name": "Bob",
"email": "bob@example.com",
"role": "user",
"active": true
},
{
"id": 3,
"name": "Cleo",
"email": "cleo@example.com",
"role": "user",
"active": false
},
{
"id": 4,
"name": "Dev",
"email": "dev@example.com",
"role": "editor",
"active": true
}
]Paste JSON, CSV, YAML or XML into the input editor, drop a file, or upload one. JSON, CSV and XML are detected automatically.
The output appears as you type. The bar under the editors shows input and output tokens, counted with a real GPT tokenizer.
Copy the result into ChatGPT, Claude or your own prompt. Ctrl/⌘ + Shift + C copies without reaching for the mouse.
TOON and TONL are most useful when you need compact, readable structured data for prompts, evaluations, and context packing. JSON remains the default for general-purpose interchange.
| Criteria | JSON | TOON / TONL |
|---|---|---|
| Best fit | General API interchange and broad tool compatibility | LLM prompts, context packing, and compact structured data |
| Syntax overhead | Repeated keys, quotes, braces, and brackets in every object | Field lists and line-oriented syntax reduce repeated structure |
| Token efficiency | Baseline for structured payloads | About 43% fewer tokens on average (about 33-59% depending on data shape) |
| Readability at scale | Familiar, but verbose on large arrays and nested objects | Compact while staying readable for tables and repeated records |
Token savings vary by input shape. Arrays of similar objects and repeated keys usually benefit the most.
A JSON to TOON converter rewrites structured data into TOON or TONL, compact text formats designed for LLM workflows. The goal is to preserve the same data while reducing repeated syntax so prompts are smaller, easier to scan, and cheaper to send.
Token savings depend on the shape of the input. Arrays of similar objects and payloads with repeated keys usually compress best. In the official TOON v4.1 benchmark, TOON used 42.6% fewer tokens than JSON on average, from about 33% on mixed or nested structures to about 59% on flat uniform tables.
No. The main web converter runs client-side in the browser, so your source data stays on your device during conversion. That makes it useful for privacy-sensitive payloads and quick experiments where you do not want to send raw data to a third party.
json2toon.co supports JSON, CSV, YAML, XML, and TOML inputs alongside TOON and TONL. You can use it to compare representations, convert existing payloads, and pick the format that fits your prompt, RAG, or configuration workflow.