Watch: cut the tokens in JSON data
No sound. Each step is captioned on screen and also written out in the guide below.
AI Tools
JSON to LLM Table
Turn JSON records into CSV or a compact table, and give an AI the same data in far fewer tokens. Exact token counts, before and after.
Shrink JSON for prompts locally
Runs locally in your browser. Plain code, no AI model. Your data is not uploaded, saved, or shared.
A list of objects, nested lists (like an API response's "data"), or an object of objects. Whole AI replies work too.
Your table will appear here.
Paste JSON records to convert them.
Tokens, exactly counted
| Format | o200k_base GPT-5, GPT-4.1, GPT-4o | cl100k_base GPT-4, GPT-3.5 | Saved vs. your JSON |
|---|
Exact counts with OpenAI's tokenizers. Claude and Gemini tokenize differently, but the savings are usually similar, because the same repeated structure is removed. For their estimates, use the AI Token Counter.
Put this line above the data
Fewer tokens in three steps
- 01Paste JSON records
API responses, database rows, logs, or any list of objects.
- 02Pick a format
The table shows exact tokens for each, so you can pick the smallest that still fits your data.
- 03Copy it into your prompt
Add the one-line description above it, so the model reads the table correctly.
Why tables are smaller
[{"orderId": "A1", "status": "shipped"},
{"orderId": "A2", "status": "pending"}]
orderId,status
A1,shipped
A2,pendingJSON repeats every field name on every row, plus quotes, braces, and indentation. A table states the names once. With dozens of rows, that often halves the tokens or better.
Frequently asked questions
Do AI models understand CSV as well as JSON?
Yes, for flat records. Models read CSV and Markdown tables reliably, especially with a one-line description above the data. For deeply nested data, or when exact types matter, use Columnar JSON, which keeps types while still listing field names only once.
What changes in CSV, TSV, and Markdown?
They store text, so numbers, true and false, and null become plain text, and null and missing values become empty cells. Nested objects become dotted columns like address.city. Columnar JSON keeps every value and type exactly.
Which format is smallest?
Usually CSV or TSV. It depends on your data, so the page counts every format and highlights the smallest.
Does this save money?
Fewer input tokens mean lower cost and more room in the context window, in proportion to the savings shown. The page shows tokens, not prices, because prices change often.
Is my data uploaded?
No. Converting and counting tokens both run in your browser.