Google DeepMind has introduced SkillSmith, a system detailed in a paper submitted to arXiv on July 29, 2026, that generates new model capabilities at inference time by blending stored text knowledge with parametric weight information, requiring no retraining. The system uses prefix-tuning to treat stored model weights as a native modality alongside traditional text input, bridging text-based knowledge composition and parametric skill libraries. SkillSmith synthesizes rich textual metadata and prefix weights together, generating new prefix weights for specific skills through an “instruction-steered paradigm.” Lead author Lucio M. Dery and six co-authors, all affiliated with Google, demonstrated that SkillSmith significantly outperforms both text-only and weight-space-only baseline models in instruction-steered tasks. While prefix-tuning is a known technique, SkillSmith adds the ability to treat prefix weights as composable building blocks that can be mixed and matched based on instructions. Discussion on social platforms like X and LinkedIn indicates noteworthy industry interest in the paper. Google’s internal collaboration, with all seven authors under one roof, highlights the coordination advantage that large labs hold over fragmented open-source or decentralized efforts.