LLM Token Counter
Free online LLM token counter. Estimate token counts for OpenAI GPT-4, Claude and Gemini models. Budget API costs and check context limits. All processing is local and private.
How to Use LLM Token Counter
Paste Your Text
Paste the text you want to count tokens for — a prompt, API response, code, or any text. The tool supports all common text input.
Select the Model
Choose the target model — GPT-4o, GPT-4, Claude Opus/Sonnet/Haiku, or Gemini Pro/Flash. Each model uses a different tokenizer, and counts will vary between them.
Read the Token Count
The tool shows estimated token count and approximate cost based on current API pricing. Use this to budget API calls, optimize prompts, or check if your input exceeds context limits.
Token Estimation Reference
| Model | Context | Est. Input Cost |
|---|---|---|
| GPT-4o | 128K | $2.50/1M tokens |
| Claude Sonnet 4 | 200K | $3.00/1M tokens |
| Gemini 2.5 Flash | 1M | $0.15/1M tokens |
| Claude Haiku 4 | 200K | $0.80/1M tokens |
Frequently Asked Questions
What is a token?
LLMs process text in tokens, not characters. A token is roughly 3/4 of a word — about 4 characters in English. "Hello world" is 2 tokens. API pricing is per token (both input and output). Context windows are measured in tokens: GPT-4o supports 128K tokens. Understanding token counts helps you budget costs and avoid hitting context limits.
Are these counts exact?
Exact tokenization requires the model's specific tokenizer (tiktoken for OpenAI, Claude's tokenizer for Anthropic). Our tool uses character-based estimation: OpenAI ~4 chars/token, Claude ~3.5 chars/token, Gemini ~4 chars/token. This is accurate within 10-15% for English text. For precise counts, use the official tokenizer libraries (tiktoken for Python, @anthropic-ai/tokenizer for Node).
Why do token counts differ between models?
Each model uses a different tokenizer with a different vocabulary. The same text can produce different token counts on GPT-4o vs Claude vs Gemini. Generally they're within 10-20% of each other for English. For code, non-English languages, and technical content, the differences can be larger.