Chrome Begins Live Testing of Its On-Device AI Engine

Google is testing LiteRT-LM for On-Device AI across Windows, Mac, and Linux via the OnDeviceModelLitertLmBackend experiment. The runtime switch is now live in a field trial for Chrome’s on-device model service.

LiteRT-LM was announced in 2025 as a production-ready engine designed to run large language models directly on devices, including Chrome. It powers certain local AI tasks in the browser, such as page summaries and writing assistance.

Chrome’s on-device AI system allows these features to run directly on your computer instead of relying fully on cloud servers. Smaller models process content locally within the browser.

Chrome has introduced AI features across the browser in recent months, including a test that prompts users to try Gemini on Android. The LiteRT-LM field trial is another step in that AI rollout.

Google is also testing LiteRT-LM behind a dedicated flag. The description reads: “Use LiteRT-LM runtime for on-device model service inference – Mac, Windows, Linux, ChromeOS.” This confirms the runtime can be manually enabled during development testing.

Google is now improving how that system runs. Recent changes improve stability and responsiveness. They improve how AI conversations are handled, ensure canceling a task reacts immediately, reduce crashes tied to AI activity, and prevent the browser from getting stuck during longer AI actions.

LiteRT-LM is now under live evaluation through Chrome’s field trial system.