Third-Party Devices Gain Gemini Home APIs and Reference Specs
Google released the Gemini built-in Program and Home APIs, allowing hardware partners to build devices with Gemini 3.5 and advanced computer vision.
On the final day of Google I/O 2026, Google made its Gemini for Home stack available to third-party hardware manufacturers and service providers. This transitions the platform from a closed ecosystem for Google Nest devices into a full-stack AI offering. Developers can now build smart home hardware utilizing the same core Gemini 3.5 models that power first-party hardware.
Hardware Reference Designs
Google introduced the Google Home Gemini built-in Program, a turnkey solution designed to reduce research and development cycles for external manufacturers. The program provides fully validated reference designs encompassing SOCs, sensors, and microphones. Google designated Amlogic as the primary system integrator, with hardware component support from SEI Robotics and Apical.
The initial rollout includes two primary hardware paths. The Camera Reference Design enables partners to build high-performance security hardware with native computer vision processing. The newly introduced Speaker Reference Design allows third-party manufacturers to build high-fidelity voice command centers capable of handling complex audio inputs.
Expanded API Capabilities
These reference designs route through the newly expanded Home APIs, driven primarily by Gemini 3.5 Flash to maintain low-latency interactions for high-frequency queries. The API expansion unlocks three specific endpoints for device manufacturers.
Camera Intelligence shifts event logging from generic motion alerts to descriptive context, allowing cameras to output specific text strings like “Julie delivered flowers” instead of basic object detection triggers. Ask Home provides a natural language query interface, giving applications the ability to answer complex questions about historical household events. Finally, the Home Brief endpoint synthesizes hours of unstructured sensor and video data into a formatted daily summary of household activity.
Carrier Integration and Subscriptions
Carriers and internet service providers can now bundle these AI capabilities directly into their own applications. AT&T serves as the initial launch partner, routing Gemini features and Nest Cam AI into its Connected Life application and security service.
To monetize the infrastructure costs of continuous video processing, Google introduced the Google Home Premium Subscription, replacing the legacy Nest Aware model. Service providers can bundle the standard tier for approximately $10 per month or $100 per year. An advanced tier is available for $20 per month. These standard capabilities are also bundled for users paying for existing Google AI Pro subscriptions at $20 per month, while the advanced tier is included with the $250 per month Ultra tier.
This release marks a structural shift from basic command-and-response smart homes toward proactive AI agents. While the program standardizes hardware development, it continues Google’s pivot toward an AI-centric strategy, sometimes at the expense of conventional Android updates. For hardware developers, building a custom smart home camera or speaker no longer requires training proprietary vision models or building custom voice synthesis pipelines. You can now rely on standardized reference hardware and Google’s routing infrastructure to handle the core perception layer.
Get Insanely Good at AI
The book for developers who want to understand how AI actually works. LLMs, prompt engineering, RAG, AI agents, and production systems.
Keep Reading
Build a Fast Multilingual OCR with Nemotron-OCR-v2
Learn how to deploy NVIDIA Nemotron-OCR-v2 for high-speed document extraction across six languages using synthetic data and GPU acceleration.
Google's Frozen v2 Chip Hardwires Gemini for 10x Efficiency
Google is designing a specialized server chip codenamed Frozen v2 that hardwires Gemini architectures directly into silicon for massive inference gains.
Photoroom-1B Filters 1 Billion Images for Product Photography
Photoroom has detailed the proprietary 1-billion-image dataset and multi-stage quality filters used to train its PRX generative AI architecture.
Vertical 60-Second Video Summaries Arrive in Google NotebookLM
Google's NotebookLM now uses the Gemini 3.1 Flash-Lite Image model to generate source-conditioned, 60-second vertical video summaries of uploaded documents.
500,000 Sensors Power Midjourney's Petaflop Ultrasonic Scanner
Midjourney Medical has unveiled a full-body ultrasonic CT scanner that processes two petaflops of data to map internal tissues in under 60 seconds.