Large language models are trained on "tokens": word fragments, rather than individual words or characters. Different LLMs can have a slightly different tokenizers: that is, they break up the same set of words into slightly different sets of tokens.
A good rule of thumb is that there's usually around three tokens for every two words, plus a few tokens of chat metadata — but that can vary based on the model, and based on your prompt's complexity. We built this calculator to help you get a sense for how many tokens a prompt uses for different open-source models.