Estimate token count and cost for text across LLM models (GPT-4o, Claude, Gemini, Mistral, Llama). Algorithmic — no API calls.