GPT Token Counter

How to use

1. 1Paste a prompt, system message, code snippet, or any text into the editor. Counts update live as you type, and a Load sample button drops in prose, JSON, code, CJK text, and number-heavy strings so you can compare token rates.
2. 2Pick a model from the sidebar. GPT-4o and GPT-4o mini use the o200k_base tokenizer; GPT-4, GPT-4 Turbo, GPT-3.5 Turbo, and text-embedding-3-large use cl100k_base. The token total adjusts automatically.
3. 3Edit the input and output dollar-per-million-token rates to match the price you actually pay (the defaults are public list-price snapshots), and set Expected output tokens to the size of the answer you expect. The total cost updates instantly.
4. 4Check the context window bar to see how much of the model's budget your prompt plus expected output consumes. The bar turns amber above 70% and red above 90%, so you know when to trim or summarize.
5. 5Use Copy summary to grab the model name, token count, byte size, context utilization, and cost estimate as plain text. Toggle Show token visualization to see each pre-token piece coloured by kind with its estimated subword token count.

About this tool

GPT Token Counter estimates how many tokens your prompt will consume on OpenAI's GPT-4o, GPT-4 Turbo, GPT-4, GPT-3.5 Turbo, and text embedding models, then turns the result into an estimated dollar cost using rates you can edit. The estimator uses the public cl100k_base and o200k_base pre-tokenization regex from OpenAI's tiktoken library to split the input into the same pieces the real tokenizer sees: contractions, runs of letters, digit groups, punctuation, and whitespace. Each piece is then scored by a calibrated subword model that mimics how the BPE merges behave in production. The result lines up within a few percent of tiktoken on plain English prose, code, and JSON, and clearly flags itself as an estimate so you can budget a small safety margin. The page also reports characters, words, UTF-8 bytes, lines, the average tokens per word ratio, and how much of the model's context window your prompt plus expected output will consume. A token visualization panel shows each pre-token chip coloured by kind (word, number, punctuation, whitespace, non-ASCII) with its estimated token count, so you can see exactly which parts of a prompt are expensive. Cost is split between input and output, using editable per-million-token prices that come pre-filled with public list-price snapshots; replace them with your contract pricing in one click. The counter is useful when sizing system prompts, budgeting few-shot examples, pricing batch workloads, planning a RAG pipeline's context budget, or simply checking whether a long document will fit before paying for the call. Nothing leaves your browser. No API key, no upload, no rate limit.

Free to use. Works in your browser. No signup, no login.