Google drops AI Edge Gallery for on-device LLMs

Google is aggressively building the "Linux of mobile AI" infrastructure by standardizing local inference workflows.

• –Features official support for Gemma 4's "Thinking Mode" and real-time on-device hardware benchmarking
• –Agent Skills augment local models with fact-grounding tools like Wikipedia and Maps without internet dependencies
• –Integrated Prompt Lab allows for granular control over parameters like temperature and sampling directly on hardware
• –Highlights a critical industry shift from cloud-reliance to "efficiency-first" mobile development paradigms
• –Community feedback notes impressive performance but flags a need for better accessibility and UX stability