Artificial intelligence is moving faster than ever, but one major challenge still slows down even the best large language models: inference latency. Google is now tackling that problem head-on with a major upgrade to Google’s Gemma 4 AI models. The company has introduced Multi-Token Prediction (MTP) drafters, a breakthrough optimization that can make Gemma 4 models generate responses up to 3x faster without sacrificing output quality, reasoning accuracy, or reliability.