Token count is an estimate based on the cl100k tokenizer heuristic (~4 chars/token). Actual count may vary by model.