Google releases Multi-Token Prediction drafters for its Gemma 4 models, which use a form of speculative decoding to guess future tokens for faster inference (Ryan Whitwam/Ars Technica)

Google's new Multi-Token Prediction drafters for Gemma 4 models significantly improve inference speed through speculative decoding, enabling organizations to deploy more efficient local AI solutions that reduce computational costs and latency. This advancement makes open-source AI models more practical for enterprise deployments, allowing IT teams to run sophisticated AI workloads on-premises with better resource utilization and faster response times. For CIOs, this represents an opportunity to reduce dependency on cloud-based AI services and improve total cost of ownership while maintaining competitive AI capabilities.