Count and visualize AI tokens across OpenAI, Anthropic Claude, Google Gemini, xAI Grok, and Llama 3. Client-side tokenization with interactive token boundary inspection.
Type or paste text on the left to see tokens
Understand your AI costs
Large-language-model providers bill per token, and every model enforces a context-window limit. Knowing the token count of a prompt or document before you send it helps you estimate cost, avoid truncation, and stay within budget.
Paste your text, choose a model family, and see the exact count instantly. Compare across providers to find the most cost-effective option for your workload.
Token boundaries are not always intuitive — a single word can split into multiple tokens, and whitespace and punctuation are encoded differently across models. The visualization highlights each token so you can see exactly how the tokenizer breaks your input apart.
Comparing the same text across OpenAI and Llama 3 encodings reveals how vocabulary differences affect token count, helping you write prompts that are efficient for the model you use.
Yes, the tool is completely free and requires no account. OpenAI and Llama 3 tokenization runs entirely in your browser using WebAssembly. Claude, Gemini, and Grok token counts are fetched through our server, which forwards your text to the provider’s counting endpoint.
For typical English prose, one word averages about 1.3 tokens. Code, JSON, URLs, non-Latin scripts, and emoji generally cost more because they contain byte sequences that don’t appear in the tokenizer’s most common vocabulary entries.
Those providers don’t publish their tokenizers publicly. They expose counting endpoints that return a total token count, but they do not return the individual token boundaries or the split itself.
The tool supports three OpenAI encodings: o200k_base (used by GPT-5.x, O1, and O3 models), cl100k_base (used by GPT-4 and GPT-3.5 Turbo), and p50k_base (used by GPT-3 and Codex). The encoding is selected automatically based on the model you choose.
For OpenAI and Llama 3, no — everything runs locally in your browser and your text never leaves your device. For Claude, Gemini, and Grok, the text is sent to our server, which forwards it to the provider’s token-counting endpoint. We do not store or log the text.
