Playground · Token Visualiser
How an LLM sees your text
Language models don't read words or letters — they read tokens: subword chunks. Common words are usually one token; rarer words split into pieces; a space normally rides along with the word that follows it. Type below and watch it happen live.
Tokens
What the model actually sees
Each token becomes an integer ID — the numbers the model really processes.
Illustrative IDs — each distinct token maps to a stable number in the range of GPT-2's 50,257-token vocabulary. A real model uses its own learned vocabulary, so the exact numbers differ.
What would this cost?
Cost to send this text 1,000 times as input, at list prices.
| Model | Input $/1M | Output $/1M | × 1,000 sends |
|---|---|---|---|
| GPT-5.5 | $5 | $30 | — |
| Claude Opus 4.7 | $5 | $25 | — |
| Claude Sonnet 4.6 | $3 | $15 | — |
| Gemini 3.5 Flash | $1.5 | $9 | — |
| DeepSeek V4 | $0.435 | $0.87 | — |
List input prices, USD per million tokens, as of May 2026. Prices change often and each provider's own tokenizer counts slightly differently, so treat these as ballpark. Output tokens are billed separately (typically 2–6× the input rate).
This is an approximation of how GPT-style tokenizers (byte-pair encoding) split text — it uses the standard pre-tokenisation step. Exact tokens and counts vary by model (GPT-4, Claude and Llama each tokenise differently). Want the real thing? Watch Karpathy build a GPT tokenizer from scratch on our Learn shelf.