Fast and lightweight token count estimation without requiring a full tokenizer.
Estimates are calibrated against OpenAI's o200k_base encoding – the tokenizer of all current GPT models. Counts for other LLM families will differ somewhat; the defaultCharsPerToken and languageConfigs options let you tune the heuristics for your model. For precise counts, use a full tokenizer like gpt-tokenizer.
- ⚡ 95%+ average accuracy, and no single sample below 90%
- 📦 Just 2kB bundle size with zero dependencies
- 🖥️ Bundled CLI – count, slice and split from the shell
- 🌍 Multi-language support with configurable language rules
- 🗣️ Built-in rules for accented scripts (German, French, Spanish, Slavic), Cyrillic, and Greek
- 🀄 CJK (Chinese, Japanese, Korean) character handling
- 😀 Emoji-aware pricing (emoji cost more tokens than their character count suggests)
The following chart shows how close the estimates come to actual GPT token counts for different input texts:
Bars grow left when tokenx underestimates and right when it overestimates. The axis spans the ±10% per-sample deviation bound.
under ◂·▸ over
Team chat transcript (en) 293 → 285 ███│ -2.73%
Vite releases API response 8,075 → 8,551 │██████ +5.89%
tokenx source code 3,151 → 3,050 ███│ -3.21%
Vite plugin API docs (en) 6,901 → 7,155 │████ +3.68%
Cat article (ja) 12,437 → 11,529 ███████│ -7.30%
Cat article (ko) 7,117 → 6,841 ████│ -3.88%
Cat article (zh) 9,057 → 8,828 ███│ -2.53%
The Great Gatsby by Fitzgerald (en) 4,391 → 4,479 │██ +2.00%
Die Verwandlung by Kafka (de) 4,437 → 4,384 █│ -1.19%
─────────────────────
mean 3.60%
Accuracy depends on the kind of text, not its length: a short excerpt deviates about as much as the full document it came from. A holdout corpus that no ratio was ever fitted against is held to a looser ±15% bound, so a retune cannot silently overfit the chart.
Three cases are knowingly outside these bounds, all underestimates:
- High-entropy strings – base64, hashes, digests: ≈-70%. Pricing them would cost every caller runtime for a case ordinary traffic rarely carries.
- Traditional and classical Chinese – the hanzi rate is calibrated on contemporary simplified script: ≈-10% to -20%.
- Scripts without a built-in rule – Arabic ≈-35%, Hindi ≈-30%, Hebrew ≈-45%, Thai ≈-60%; a custom language rule closes the gap.
# npm
npm install tokenx
# pnpm
pnpm add tokenx
# yarn
yarn add tokenxThe package ships a tokenx binary – no install needed via npx, or install it for the commands below.
# Count tokens in a file, a pipe, or several files at once
tokenx README.md
cat article.md | tokenx
tokenx count src/*.ts
# Fail a script when a prompt outgrows its budget (exit code 2)
tokenx prompt.txt --limit 8000
# Extract a token range, or chunk a document for RAG
tokenx slice article.md --end 500
tokenx split article.md --size 500 --overlap 50Only results go to stdout – counts as bare integers, chunks as a JSON array – so $(tokenx count file.txt) and | jq work unchanged. Notices and errors go to stderr. Run tokenx --help for every option.
import { estimateTokenCount, isWithinTokenLimit, sliceByTokens, splitByTokens } from 'tokenx'
const text = 'Your text goes here.'
// Estimate the number of tokens in the text
const estimatedTokens = estimateTokenCount(text)
console.log(`Estimated token count: ${estimatedTokens}`)
// Check if text is within a specific token limit
const tokenLimit = 1024
const withinLimit = isWithinTokenLimit(text, tokenLimit)
console.log(`Is within token limit: ${withinLimit}`)
// Slice text by token positions (like Array.slice)
const firstTokens = sliceByTokens(text, 0, 5)
console.log(`First ~5 tokens: ${firstTokens}`)
// Split text into token-based chunks
const chunks = splitByTokens(text, 100)
console.log(`Split into ${chunks.length} chunks`)
// Use custom options for different languages or models.
// Custom language rules are checked before all built-in heuristics,
// so they can also override the built-in CJK handling.
const customOptions = {
defaultCharsPerToken: 4, // More conservative estimation
languageConfigs: [
{ pattern: /[\u4E00-\u9FFF]/, averageCharsPerToken: 2 }, // Custom Chinese rule
]
}
const customEstimate = estimateTokenCount(text, customOptions)
console.log(`Custom estimate: ${customEstimate}`)Estimates the number of tokens in a given input string using heuristic rules that work across multiple languages and text types.
Usage:
const estimatedTokens = estimateTokenCount('Hello, world!')
// With custom options
const customEstimate = estimateTokenCount('Bonjour le monde!', {
defaultCharsPerToken: 4,
languageConfigs: [
{ pattern: /[éèêëàâîï]/i, averageCharsPerToken: 3 }
]
})Type Declaration:
function estimateTokenCount(
text?: string,
options?: TokenEstimationOptions
): number
interface TokenEstimationOptions {
/** Default average characters per token when no language-specific rule applies (default: 7). */
defaultCharsPerToken?: number
/** Custom language configurations to override defaults. */
languageConfigs?: LanguageConfig[]
}
interface LanguageConfig {
/** Regular expression to detect the language. */
pattern: RegExp
averageCharsPerToken: number
}Checks if the estimated token count of the input is within a specified token limit.
Usage:
const withinLimit = isWithinTokenLimit('Check this text against a limit', 100)
// With custom options
const customCheck = isWithinTokenLimit('Text', 50, { defaultCharsPerToken: 3 })Type Declaration:
function isWithinTokenLimit(
text: string,
tokenLimit: number,
options?: TokenEstimationOptions
): booleanExtracts a portion of text based on token positions, similar to Array.prototype.slice(). Supports both positive and negative indices.
Usage:
const text = 'Hello, world! This is a test sentence.'
const firstThree = sliceByTokens(text, 0, 3)
const fromSecond = sliceByTokens(text, 2)
const lastTwo = sliceByTokens(text, -2)
const middle = sliceByTokens(text, 1, -1)Type Declaration:
function sliceByTokens(
text: string,
start?: number,
end?: number,
options?: TokenEstimationOptions
): stringParameters:
text- The input text to slicestart- The start token index (inclusive). If negative, treated as offset from end. Default:0end- The end token index (exclusive). If negative, treated as offset from end. If omitted, slices to the endoptions- Token estimation options (same asestimateTokenCount)
Returns:
The sliced text portion corresponding to the specified token range.
Splits text into chunks based on token count. Useful for chunking documents for RAG, batch processing, or staying within context windows.
tokensPerChunk is a target, not a hard maximum: a chunk closes once it reaches the target, so a single long segment can push a chunk beyond it. Chunks never break words apart. Segments split on whitespace and punctuation, so for CJK text – where a whole clause between two punctuation marks is a single segment – chunks can far exceed the target.
Usage:
const text = 'Long text that needs to be split into smaller chunks...'
// Basic splitting
const chunks = splitByTokens(text, 100)
console.log(`Split into ${chunks.length} chunks`)
// With overlap for semantic continuity
const overlappedChunks = splitByTokens(text, 100, { overlap: 10 })
// With custom options
const customChunks = splitByTokens(text, 50, {
defaultCharsPerToken: 4,
overlap: 5
})Type Declaration:
interface SplitByTokensOptions extends TokenEstimationOptions {
/** Number of tokens to overlap between consecutive chunks (default: 0, clamped below `tokensPerChunk`). */
overlap?: number
}
function splitByTokens(
text: string,
tokensPerChunk: number,
options?: SplitByTokensOptions
): string[]Parameters:
text- The input text to splittokensPerChunk- Target number of tokens per chunkoptions- Token estimation options with optional overlap
Returns:
An array of text chunks, each containing approximately tokensPerChunk tokens. With overlap, each chunk repeats the trailing tokens of the previous one; a final chunk consisting only of overlap content is never emitted.
MIT License © 2023-PRESENT Johann Schopplich