Token Counter

Visualise how AI models split text into tokens

What is a token? LLMs don't read words — they read tokens: fragments of roughly 3–4 characters. A word like tokenization becomes three tokens: token + iz + ation. Tokens determine both latency and cost — every API call bills per token.(Approximation of cl100k_base, the encoding used by GPT-4 and Claude.)

Tokens

Words

Chars

Token breakdown(each colour = one token)

The·quick·brown·fox·jumps·over·the·lazy·dog.

Middle dots (·) represent spaces. Hover a token to see its index and character count.

ModelInput /1MOutput /1MThis inputEst. output×3

GPT-4oOpenAI

$2.5$10$0.000025$0.000325

GPT-4o miniOpenAI

$0.15$0.6$0.000002$0.000019

Claude Sonnet 4Anthropic

$3$15$0.000030$0.000480

Claude Haiku 3.5Anthropic

$0.8$4$0.000008$0.000128

Gemini 1.5 ProGoogle

$1.25$5$0.000013$0.000163

Gemini 1.5 FlashGoogle

$0.075$0.3<$0.000001$0.000010

"Est. output×3" assumes the model generates 3× the input tokens — adjust for your use case. Prices approximate as of mid-2025.

~4 chars per token

Rough rule for English prose. Technical jargon, code, and non-English text often use more tokens per word.

Spaces are part of tokens

In GPT-4 / Claude encoding, a leading space is merged with the following word — " hello" is one token, not two.

Context window = token budget

Models have a fixed context window (e.g. 200K tokens for Claude). Your prompt + history + response must all fit.