Estimate how many tokens your prompt uses across major AI models. Based on the ~4 chars/token approximation used by OpenAI and Anthropic.