Token Counter & Visualizer

Paste any text and see how each model family tokenizes it. GPT counts are exact (tiktoken runs in your browser); Claude, Gemini, and Llama use calibrated estimates. Nothing is uploaded.

0 characters 0 words

Token visualization (GPT-4o / GPT-5)

Each colored chunk is one token (o200k_base).

How this works

GPT-4o and GPT-5 use the o200k_base encoding; GPT-4 and GPT-3.5 use cl100k_base. Both are tokenized exactly here using gpt-tokenizer, a pure-JavaScript port of OpenAI's tiktoken — it runs entirely in your browser, so your text is never sent anywhere.

Claude, Gemini, and Llama don't publish a JavaScript tokenizer, so their counts here are calibrated estimates based on each family's average characters-per-token ratio. They're close enough to compare relative cost, but for exact billing use each vendor's own tokenizer or the usage field on a real API response.

Why token counts differ between models

Each model family trains its own byte-pair-encoding vocabulary, so the same sentence can split into a different number of tokens. Anthropic's newest tokenizer (Claude Opus 4.7) uses up to ~35% more tokens for the same English text than older encodings — which is why "cheaper per token" doesn't always mean "cheaper per request." To turn token counts into dollars, use our LLM Cost Calculator.