← Home
中文

Token Counter for LLMs

Estimate GPT-4, GPT-3.5 and Claude token usage and cost

Est. Tokens
0
approximate count
Characters
0
0 without spaces
Words
0
English + CJK
Char/Token
—
average ratio

💰 Estimated API cost (per request, input only)

About this estimator

  • For accurate tokenization, OpenAI/Anthropic require their own tokenizer libraries. This tool uses heuristics calibrated to match real outputs within ~10-15% for typical text
  • English text: ~4 characters per token on average for cl100k_base (GPT-4)
  • Chinese/Japanese/Korean: ~1.5-2 characters per token (CJK takes more tokens per char)
  • Code, JSON and structured text typically use more tokens than prose due to special characters

Use cases

  • Estimate API cost before sending a prompt
  • Stay within model context windows (4K/8K/32K/128K/200K)
  • Trim prompts to reduce latency and cost
  • Compare token efficiency between GPT and Claude for the same content