Apple’s latest Gemma 4 model introduces Multi-Token Prediction drafters to sharpen inference speed. By combining a powerful target model with a specialized drafter, the updates deliver up to three times faster token generation without sacrificing quality. This advancement addresses a major bottleneck in LLM inference, especially for developers relying on real-time processing.

The tech promises sharper responsiveness for apps, agents, and cloud services, making AI more practical on edge devices. Engineers now see tangible gains in performance, positioning Gemma 4 as a stronger choice for developers seeking efficiency.