Type or paste anything below and watch how three real tokenizers carve it up. The same text becomes wildly different token sequences depending on which model is processing it. Cost, context length, and prompt sensitivity all live downstream of this.
Add an open-weight tokenizer for cross-vendor comparison? We'll try Claude, Mistral, Qwen, Phi-3, Gemma, and Llama 2 in order until one loads (~1–5 MB depending on which).