Official2026-06-30

DeepSeek Publishes Token Heuristics and Offline Tokenizer

Checked on June 30, 2026: DeepSeek's official Token & Token Usage page says tokens are the billing unit, gives rough heuristics of about 0.3 token per English character and 0.6 token per Chinese character, and links a `deepseek_tokenizer.zip` package for offline token estimation while still treating API `usage` output as the real source of truth.

What we verified on June 30, 2026

  • DeepSeek's official Token & Token Usage page defines tokens as the basic unit for both text representation and billing.
  • The same page publishes simple estimation heuristics: about 0.3 token per English character and about 0.6 token per Chinese character.
  • DeepSeek also warns that conversion ratios vary by model tokenization, so the real count for any request should come from the API's returned usage data rather than from the rough heuristic alone.
  • The page now links an offline package named deepseek_tokenizer.zip for local input/output token estimation outside the live API response path.
  • This is practical infrastructure guidance rather than storefront guidance: it helps developers estimate spend and prompt size, but it does not create any new /pricing inventory or plan card.

Sources

View original source