{
  "id": 8341792,
  "title": "How many tokens is 1,000 words? A conversion cheat sheet for LLM prompts",
  "url": "https://urgent.news/2026/09/19/how-many-tokens-is-1-000-words-a-conversion-cheat-sheet-for-llm",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-19T00:19:02.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/ilostcount/how-many-tokens-is-1000-words-a-conversion-cheat-sheet-for-llm-prompts-3n17"
  },
  "original_language": "en",
  "account": "1,000 words of standard English equates to roughly 1,300 tokens. Conversely, 1,000 tokens translates to about 750 words or approximately 4,000 characters. This forms the core answer for calculating token counts. The remainder of this article provides a table, instances where the ratio deviates, and methods to obtain precise counts when estimates are insufficient. The Cheat Sheet In everyday English, the common estimation suggests 1 token corresponds to about 4 characters or 0.75 words:\n\n* One word translates to roughly 1.3 tokens\n* 100 words are approximately 130 tokens\n* A page of 500 words equals about 650 tokens\n* 1,000 words amount to around 1,300 tokens\n* Ten pages approximate 6,500 tokens\n* 1,000 characters equal about 250 tokens\n* One character is roughly 0.25 tokens\n* One paragraph, consisting of around 100 words, is approximately 130 tokens\n\nWhen dealing with a model's context limit, you typically need the number of tokens in reverse. A token budget of 1,000 tokens corresponds to approximately 750 words. For 4,000 tokens, that is close to 3,000 words. Similarly, 8,000 tokens equate to around 6,000 words, and 128,000 tokens are roughly equivalent to 96,000 words (equivalent to a short novel). The ratio breaks down in certain situations. These figures pertain to prose. Tokenizers divide on statistical frequency rather than words, meaning unusual content, such as code, JSON, and lengthy URLs, breaks into numerous small pieces, necessitating additional token usage. Punctuation, braces, camelCase, and random ID strings all tokenize inefficiently. Consequently, a budget exceeding the standard prose ratio is required. Non-Latin scripts, including Arabic, Chinese, Japanese, Hindi, and others, often translate to multiple tokens per character in some tokenizers, causing the same meaning to require more tokens than English. Numerical data and tables tokenize in arbitrary ways. A lengthy column of figures does not incur the same cost as text. Different models and tokenizers yield varying token counts. A prompt is not a static token count but varies based on the tokenizer. Therefore, while rules of thumb are suitable for estimating if the prompt will fit, they are not reliable for hard limits or cost calculations dependent on accurate estimates. Determining an exact count through code For OpenAI models, the tiktoken library facilitates direct token counting:\n\nimport tiktoken\nenc = tiktoken.get_encoding(\"o200k_base\")\nprint(len(enc.encode(\"your prompt here\")))\n\nAnthropic and Google also provide token-counting endpoints in their APIs, offering the most accurate method when estimating the count for the exact model intended for use instead of approximations. Determining an exact count without programming Most of the time, you are not programming. You have a text block in front of you and wish to ascertain whether it fits before incorporating it. This need is addressed by iLostCount:\n\nPaste the text, and the tool instantly displays the token, word, and character counts. No registration is required, and the counting occurs directly on the webpage, meaning the text is never uploaded or stored. Knowing the number holds significance as the context window is shared among the system prompt, previous conversation, any retrieved information, and the response. Overfilling the input results in no room for the output. Every token, both input and output, incurs a charge. Truncation often occurs silently, leading to a gradual increase in costs as the model \"forgets\" content due to truncated older turns. This can create the illusion that the model is forgetting when, in fact, the input limit was exceeded. By counting a lengthy document before including it in a prompt, you can avoid such issues. For example, if a 40-page PDF converts to 30,000 tokens, you would know to segment the content or summarize it, preventing unexpected errors or inflated invoices. Disclosure: this article originates from the iLostCount project. The tool is free, and its source code is publicly available at github.com/ahmad-almazeedi/token-counter.",
  "summary": "If you only want the number: 1,000 words of ordinary English is roughly 1,300 tokens. Going the other way, 1,000 tokens is roughly 750 words, or about 4,000 characters. That is the whole answer for estimating. The rest of this post is the table, the cases where the ratio breaks, and how to get an exact count when an estimate is not good enough. The cheat sheet For plain English prose, using the…",
  "key_points": [
    "1,000 standard English words equal roughly 1,300 tokens",
    "1,000 tokens correspond to about 750 words or 4,000 characters",
    "Token counts vary for non-Latin scripts and numerical data"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}