Verified October 2026

Home › Tools › Tokens per word: Japanese

Tokens per word in Japanese

Japanese averages about 1.8 tokens per character on current BPE tokenizers, which is 35% more tokens than the same count in English. Counted per character. Mixed kana and kanji plus no spaces makes token density higher than Chinese.

1.8Tokens per character
0.56characters per token
1,800Tokens per 1,000 characters
+35%vs English

Japanese to tokens table

Characters → Tokens (Japanese, 1.8 tokens per character)
CharactersTokens
100180
500900
1,0001,800
5,0009,000
10,00018,000
50,00090,000
100,000180,000

Language comparison

LanguageTokens per characterTokens for 1,000Cost multiplier vs English
English1.331,3301.00×
Spanish1.71,7001.28×
French1.81,8001.35×
German1.91,9001.43×
Portuguese1.71,7001.28×
Russian2.42,4001.80×
Arabic2.92,9002.18×
Hindi3.23,2002.41×
Chinese1.51,5001.13×
Japanese1.81,8001.35×

Ratios are per word except Chinese and Japanese, which are per character. Comparing them directly is approximate — the honest way to budget is to tokenize a sample of your own corpus.

Cost of 1,000 characters in Japanese

Cost of 1,800 tokens (1,000 characters) — USD, rates verified October 2026
#ModelProviderInputOutputBlended
1GPT-6 LunaOpenAI$0.00018$0.0009$0.0004
2GPT-5.6 LunaOpenAI$0.00036$0.00216$0.0009
3GPT-5.4 nanoOpenAI$0.00036$0.00225$0.00092
4Gemini 3.5 Flash-LiteGoogle Gemini$0.00054$0.0045$0.00166
5DeepSeek V4.1 FlashDeepSeek$0.00054$0.00216$0.00108
6GPT-5.4 miniOpenAI$0.00135$0.0081$0.00337
7Gemini 3.8 FlashGoogle Gemini$0.00135$0.00675$0.00304
8Gemini 3.7 FlashGoogle Gemini$0.00135$0.00675$0.00304
9Claude Haiku 4.5Anthropic$0.0018$0.009$0.00405
10Gemini 2.5 ProGoogle Gemini$0.00225$0.018$0.00675
11Grok 4.3xAI Grok$0.00225$0.0045$0.00337
12DeepSeek V4 ProDeepSeek$0.00238$0.00713$0.00416
13Gemini 3.5 FlashGoogle Gemini$0.0027$0.0162$0.00675
14GPT-5.2OpenAI$0.00315$0.0252$0.00945
15GPT-6.1 SolOpenAI$0.0036$0.018$0.0081
16GPT-6 SolOpenAI$0.0036$0.018$0.0081
17GPT-5.6 TerraOpenAI$0.0036$0.0216$0.009
18GPT-4.1OpenAI$0.0036$0.0144$0.0072
19Claude Sonnet 5Anthropic$0.0036$0.018$0.0081
20Gemini 3.1 Pro (Preview)Google Gemini$0.0036$0.0216$0.009
21Grok 4.6xAI Grok$0.0036$0.0108$0.0063
22GPT-5.4OpenAI$0.0045$0.027$0.0112
23Claude Sonnet 4.6Anthropic$0.0054$0.027$0.0122
24GPT-5.6 SolOpenAI$0.0072$0.036$0.0162
25GPT-5.5OpenAI$0.009$0.054$0.0225
26Claude Opus 5Anthropic$0.009$0.045$0.0202
27GPT-6 AstraOpenAI$0.018$0.09$0.0405
28Claude Fable 5.1Anthropic$0.018$0.09$0.0405

Input = 1,800 prompt tokens. Output = 1,800 generated tokens. Blended = input + 25% of output, the same weighting used across this site.

What this means for a multilingual product

If you serve Japanese and English from the same budget, Japanese consumes 1.4× the tokens for the same number of characters. Either the price per token absorbs that, or the Japanese product quietly runs at a lower margin. Measuring the ratio on your own text is the only reliable way to price it.

FAQ

How many tokens is 1,000 characters in Japanese?

About 1,800 tokens, using 1.8 tokens per character. That is 35% more tokens than the same count in English (1,330).

Why is Japanese more expensive per character?

Counted per character. Mixed kana and kanji plus no spaces makes token density higher than Chinese. Providers bill per token, not per character, so a language that fragments into more tokens costs more for the same meaning.

How do I get an exact count for Japanese?

Run a sample of your own text through the tokenizer of the model you will be billed on. Public ratios are averages; product names, numbers and formatting in your own data can move the result by 20% or more.

Keep going

Calculate a real monthly bill