Cheapest per token ≠ cheapest per Thai text
Today the cheapest model per token (Qwen3.7 Flash) is also the cheapest per Thai text, but the rest of the order changes once you count characters.
Why Thai costs more
The same article costs 1.67× the tokens in Thai as in English on GPT; across vendors the range is 1.20–3.47×. The last column shows each model’s figure.
How it is calculated
- Price per million characters = that model’s Thai tokens per character × its price per million tokens (a million characters is about 120 news articles). Input and output use the same rate — it is the same tokenizer.
- Thai tokens per character were measured on 109 real Thai articles on this site (tech news with English terms mixed in). GPT, Qwen and Typhoon use their real tokenizers; the others were measured through OpenRouter.
- Prices come from OpenRouter and the exchange rate from Frankfurter on every site rebuild (daily). Cache and batch discounts and other providers’ prices are not included.
Related tools: Count tokens in real text · How much Thai fits in each model