This is called Speculative Decoding. It speeds up LLMs by ~100%. References: OpenAI: https://openai.com/index/gpt-5-6-frontier-intelligence-efficiency/ Deepseek: https://arxiv.org/pdf/2607.05147 Gemini: https://research.google/blog/looking-back-at-speculative-decoding/ AI Engineering Website: htt...