OpenAI Current tokenizer sourceHigh confidenceChecked 2026-07-19
JSON is encoded locally with o200k_base; no chat or API wrapper tokens are added.
Paste a JSON payload, compare token counts, and see how formatting or minification changes the result.
Use valid JSON, then format or minify it before running the same payload through seven local tokenizers.
Tokenizer files load only after you run a comparison. Your text stays in this browser.
| Tokenizer | Tokens | Tokens / 100 characters | Difference from OpenAI Current |
|---|---|---|---|
| OpenAI Currento200k_base | 37 | 32.7 | Same as baseline |
| OpenAI Legacycl100k_base | 37 | 32.7 | Same as baseline |
| Gemma 3Pinned SentencePiece model | 45 | 39.8 | +8 (+21.6%) |
| DeepSeek V4 ProPinned tokenizer JSON | 40 | 35.4 | +3 (+8.1%) |
| Qwen 3.5Pinned tokenizer JSON | 43 | 38.1 | +6 (+16.2%) |
| Mistral Medium 3.5Pinned tokenizer JSON | 42 | 37.2 | +5 (+13.5%) |
| Llama 3Pinned tokenizer JSON | 37 | 32.7 | Same as baseline |
The result includes every key, value, quote, comma, brace, space, and line break in the editor. It does not include the surrounding API message, tool schema, or provider-specific formatting.
Compare ordinary text in the multi-tokenizer comparison, or inspect individual token boundaries in the token counter.
OpenAI Current tokenizer sourceHigh confidenceChecked 2026-07-19
JSON is encoded locally with o200k_base; no chat or API wrapper tokens are added.
OpenAI Legacy tokenizer sourceHigh confidenceChecked 2026-07-19
JSON is encoded locally with cl100k_base; no chat or API wrapper tokens are added.
Gemma 3 tokenizer sourceHigh confidenceChecked 2026-07-19
JSON is encoded locally with Pinned SentencePiece model; no chat or API wrapper tokens are added.
DeepSeek V4 Pro tokenizer sourceHigh confidenceChecked 2026-07-19
JSON is encoded locally with Pinned tokenizer JSON; no chat or API wrapper tokens are added.
Qwen 3.5 tokenizer sourceHigh confidenceChecked 2026-07-19
JSON is encoded locally with Pinned tokenizer JSON; no chat or API wrapper tokens are added.
Mistral Medium 3.5 tokenizer sourceHigh confidenceChecked 2026-07-19
JSON is encoded locally with Pinned tokenizer JSON; no chat or API wrapper tokens are added.
Llama 3 tokenizer sourceMedium confidenceChecked 2026-07-19
JSON is encoded locally with Pinned tokenizer JSON; no chat or API wrapper tokens are added.
It can. Spaces, indentation, and line breaks are part of the encoded text. Use the format and minify controls to compare both representations.
No. JSON parsing and tokenization run in your browser. Do not paste secrets into any online tool even when local processing is available.
No. It counts the exact JSON string in the editor. Chat roles, tool definitions, and provider request formatting are not included.
The page requires valid JSON before running the comparison so formatting mistakes are not mistaken for payload data.