Token counter
Multi-model AI token counter
Count tokens for 32 frontier models across GPT, Claude, Gemini, and Llama. Runs entirely in your browser. Nothing is uploaded.
Cost estimate assumes 1,000 calls of this prompt as input plus 200 output tokens per call. Tokenizer for non-OpenAI models is approximated with cl100k_base or char/4, typically within 5 to 15% of true count. Claude models from Opus 4.7 onward (including Opus 5, Sonnet 5, and the Fable family) use a newer tokenizer that produces roughly 30% more tokens for the same text, so counts for those models may run low. Pricing data last verified 2026-09-15.
What a token actually is
Tokens are the units a language model "sees." Each model has its own tokenizer, a function that splits text into pieces. A 100-character prompt might be 25 tokens for GPT-4o and 30 for Llama, depending on the encoding.
How accurate are these counts
For OpenAI models we use the official o200k_base tokenizer
(the same one OpenAI's API uses), so counts are exact.
For Claude, Gemini, Llama, and Grok, the official tokenizers aren't easily runnable in a
browser. We approximate using cl100k_base for
Claude and Llama, o200k_base for Grok
(typically within 5 to 10% of true count), and a character-based estimate for Gemini.
For exact counts on these, use each provider's own count-tokens API.
About the cost estimate
The cost estimate assumes 1,000 calls of your prompt as input plus 200 output tokens per call, priced at each model's input/output rate. Tweak this assumption to fit your actual workload on the cost calculator.