Free tool · Tokens
Token counter
Paste any text and count its tokens with OpenAI's o200k_base tokenizer, then see what it costs as input or output on all 19 models in llmwise. Free, in your browser: the text isn't sent anywhere.
…
tokens, OpenAI o200k_base
Exact for that tokenizer (GPT-4o, GPT-4.1, the o-series)
97
tokens, llmwise's estimate
Characters ÷ 4, the rule llmwise uses for long chats
386
characters
71
words
Sent in llmwise, a message counts as 1 on your plan whatever its length, until the chat passes 64,000 tokens.
What this text costs on each model's API
At each model's catalog price per million tokens, from the o200k_base count: an estimate for every model below, since each counts with its own tokenizer.
| Model | Sent as input | Written as output | Fits its context |
|---|---|---|---|
| Claude Fable 5.1Anthropic | $0.00097 | $0.0049 | Yes (1,000,000) |
| Claude Opus 5.5Anthropic | $0.00039 | $0.0019 | Yes (1,000,000) |
| Claude Sonnet 5.5Anthropic | $0.00019 | $0.00097 | Yes (1,000,000) |
| Claude Sonnet 5Anthropic | $0.00019 | $0.00097 | Yes (1,000,000) |
| Claude Haiku 5.5Anthropic | $0.0000097 | $0.000048 | Yes (1,000,000) |
| Claude Haiku 4.5Anthropic | $0.000097 | $0.00049 | Yes (200,000) |
| GPT-6 AstraOpenAI | $0.00097 | $0.0049 | Yes (1,050,000) |
| GPT-6.1 SolOpenAI | $0.00019 | $0.00097 | Yes (1,050,000) |
| GPT-6 SolOpenAI | $0.00019 | $0.00097 | Yes (1,050,000) |
| GPT-6 LunaOpenAI | $0.0000097 | $0.000048 | Yes (1,050,000) |
| Gemini 3.1 ProGoogle | $0.00019 | $0.0012 | Yes (1,048,576) |
| Gemini 3.8 FlashGoogle | $0.000073 | $0.00036 | Yes (1,048,576) |
| DeepSeek V4.1 FlashDeepSeek | $0.000029 | $0.00012 | Yes (1,048,576) |
| DeepSeek V4 ProDeepSeek | $0.000039 | $0.00039 | Yes (1,048,576) |
| Grok 4.7xAI | $0.00019 | $0.00058 | Yes (500,000) |
| Kimi K3Moonshot | $0.00029 | $0.0015 | Yes (1,048,576) |
| GLM 5.3Z.ai | $0.00014 | $0.00043 | Yes (1,048,576) |
| GLM 5.3 FlashZ.ai | $0.000015 | $0.000048 | Yes (1,048,576) |
| Mistral Large 4Mistral | $0.000066 | $0.00020 | Yes (1,048,576) |
Counted in your browser: your text isn't sent anywhere. The tokenizer loads the first time the page counts.
Prices checked . Updated .
How many tokens is a word? Four samples
The same short message, counted four ways: OpenAI's o200k_base tokenizer, which the counter above uses, and llmwise's own estimate of 4 characters a token.
| Sample | Words | Characters | Tokens (o200k_base) | Tokens a word | llmwise's estimate |
|---|---|---|---|---|---|
| An English email | 71 | 386 | 82 | 1.2 | 97 |
| Python code | 25 | 247 | 62 | 2.5 | 62 |
| Spanish | 44 | 228 | 51 | 1.2 | 57 |
| Hindi | 44 | 201 | 63 | 1.4 | 51 |
Plain English runs at about 1.2 tokens a word; code at 2.5, because every bracket, quote and indent is a token of its own. llmwise's rule of thumb (4 characters a token) gave 97 for the English sample against 82 counted, and 51 for the Hindi one against 63: close enough to decide when a chat is long, not a count to bill by.
Why every model counts differently
OpenAI publishes its tokenizers (tiktoken). o200k_base is the one it lists for its models from GPT-4o on, the o-series included; it hasn't said which one GPT-6 uses.
Anthropic (Claude), Google (Gemini) and xAI (Grok) count tokens with their own tokenizers, reachable through their APIs but not published to run in a browser.
The open-weight models in llmwise (DeepSeek, Kimi, GLM) ship their tokenizers with their weights; they're large files, so this page doesn't load them.
So the o200k_base figure is exact for that tokenizer, and for every model in llmwise the costs above are an estimate built on it: a model's own count can differ from it.
Tokens in llmwise: you pay per message
llmwise doesn't bill tokens. A message counts as one on your plan (or a few, on the biggest models), and the message box shows what it counts as before you send. Tokens matter in one place: a chat's length. Past 64,000 tokens a reply counts as 2 messages, past 128,000 as 4, and a chat holds up to 200,000. The message calculator shows what that means on each plan, and the LLM cost calculator compares a month of API calls with a plan.
Questions
How many tokens is 1,000 words?
About 1,155 tokens by OpenAI's o200k_base tokenizer, going by our English sample (71 words, 82 tokens: 1.2 a word). Other models' tokenizers give a somewhat different number, and code or other languages take more tokens a word.
Is the count exact for Claude, Gemini or GPT-6?
No. The count here is exact only for OpenAI's o200k_base tokenizer. Anthropic, Google and xAI count tokens through their own APIs and don't publish a tokenizer you can run in a browser, so for Claude, Gemini and Grok it's an estimate; the same goes for GPT-6, since OpenAI hasn't said which tokenizer it uses.
Do spaces and punctuation count as tokens?
Yes. A token is a piece of text the model reads: a word, part of a word, a space joined to the word after it, or a punctuation mark. Line breaks and indentation count too, which is why code takes more tokens a word.
Why does Hindi (or Japanese) take more tokens?
Tokenizers are built mostly from English text, so other scripts split into more pieces: our Hindi sample took 1.4 tokens a word against 1.2 for the same message in English.
How much do a million tokens cost?
Sent as input, from $0.10 on Claude Haiku 5.5 to $10.00 on GPT-6 Astra, at the catalog prices of the 19 models in llmwise. Written as output it costs more on every one of them. In llmwise you don't pay per token: a message counts as one on your plan.
Claude, GPT, Gemini, DeepSeek, Grok, Kimi, GLM, and Mistral, in one chat.
See what a message costs before you send it. Free is 5 messages to try; sign in with an email link, no password or card.