Paste a prompt to count its tokens. Exact counts for OpenAI models (loaded on demand, entirely in your browser); estimated counts for other models.
OpenAI switched to a new, larger vocabulary (o200k_base) for GPT-4o, GPT-4.1 and o1 — it's more efficient per token but incompatible with the older cl100k_base tokenizer used by GPT-3.5-turbo, GPT-4 and GPT-4-turbo. Pick the right option above for an exact count.
No. Tokenizing happens entirely in your browser after a one-time download of the tokenizer data. Nothing you type is sent to a server.
Anthropic and Google don't publish a client-side tokenizer library — getting an exact count would require sending your text to their API. The estimate here uses a character-based heuristic that's usually within a few percent.