ℹ️ This is an estimate, not an exact count. Real tokenizers (like OpenAI's tiktoken or Anthropic's) use model-specific vocabularies that this simple heuristic can't replicate exactly. It's accurate enough for gut-checking prompt length and rough cost estimates, but treat the exact number as approximate — usually within about 10–15% of the real count for English text.
Approximate cost per model (input tokens only, USD)