Kimi Token Saverby Faris Dzakialie Rusputra$5Reduce LLM input tokens and costs by rewriting prompts for maximum density and requirement fidelity.10