Online tokenizer

Count tokens online by AI model

Paste text into the tokenizer family that matches your model. Get a real token count, inspect token pieces and IDs when available, and understand when a plain-text result will differ from an API total.

Runs locally · Token pieces + IDs

OpenAI Tokenizer

Count plain-text tokens with current or legacy OpenAI encodings, then inspect each token boundary and ID in your browser.

Open tokenizer Runs locally · Token pieces + IDs

Gemini Tokenizer

Count Gemini-compatible plain-text tokens and inspect token boundaries and IDs in your browser.

Open tokenizer Official API count · API input total

Claude Token Counter

Count the API input tokens for a Claude user message with Anthropic's official token-count endpoint.

Open tokenizer Runs locally · Token pieces + IDs

Llama Tokenizer

See how a Llama 3 tokenizer splits plain text, inspect token IDs, and keep your text in the browser.

Open tokenizer Runs locally · Token pieces + IDs

Gemma Tokenizer

See how Gemma 3 splits plain text into tokens, inspect token IDs, and keep the text in your browser.

Open tokenizer Runs locally · Token pieces + IDs

Qwen Tokenizer

Tokenize text with the pinned Qwen 3.5 tokenizer, inspect token boundaries and IDs, and keep your input local.

Open tokenizer Runs locally · Token pieces + IDs

Mistral Tokenizer

Paste text, a prompt, or code to count tokens with the pinned Mistral Medium 3.5 tokenizer, then inspect the exact token pieces and IDs in your browser.

Open tokenizer Runs locally · Token pieces + IDs

DeepSeek Tokenizer

Count tokens with the pinned DeepSeek V4 Pro tokenizer and inspect the token pieces and IDs without uploading your text.

Open tokenizer

Choose the right count

Tokenizer results depend on the model family

There is no universal token count for the same text. Use the matching tokenizer when you need a model-specific input total.

Local tokenizers

OpenAI, Gemini, Gemma, Llama, Qwen, Mistral, and DeepSeek

These pages run their pinned tokenizer assets in your browser and show the produced pieces and IDs. StackLens does not upload the text.

Claude

Official API input count

The Claude option uses Anthropic's token-count endpoint, so the submitted text is sent to Anthropic for counting. StackLens does not store it.

Plain text boundary

Count first, estimate second

Roles, tools, chat templates, images, and retries can change an API bill. Use the measured count in the API cost calculator for scenario planning.

All families

Use the main token counter

Switch between tokenizer families without leaving one workspace.

Same input

Compare tokenizers

Run the same text through seven local tokenizers.

Documents

Count tokens in files

Count text, Markdown, JSON, CSV, and code files locally.

FAQ

Online tokenizer questions

Which online tokenizer should I choose?

Choose the model family that will process the text. Token boundaries and IDs are specific to a tokenizer, so the same text can have a different count in another family.

Do these tokenizers upload my text?

OpenAI, Gemini, Llama, Gemma, Qwen, Mistral, and DeepSeek tokenization runs locally in your browser. Claude uses Anthropic's official token-count API and sends the submitted text to Anthropic for counting; StackLens does not store it.

Will a plain-text tokenizer match an API bill exactly?

Not always. Roles, chat templates, tool definitions, images, and provider-specific request formatting can add tokens beyond the plain text entered into a tokenizer.