Google unveiled its most advanced generative AI model, Gemma 3n, to the world at the Google I/O 2025 event on Tuesday. In addition, this new model is built to seamlessly work across all devices — mobile, desktop, and tablet. It even runs well on devices with under 2GB of RAM. This accessibility minimizes barriers for users who may not have access to higher-specification hardware.
Gemma 3n is nearly identical in architecture to the Gemini Nano, which had been released earlier this summer. It’s designed to provide astonishing generation across all modalities—audio, text, images and video. Google’s pitch for this model has been that, if done right, it’s a better experience for the user because it runs seamlessly without requiring a ton of device resources.
Along with Gemma 3n, Google shared news of MedGemma’s release through its Health AI Developer Foundations program. MedGemma is being marketed as Google’s most powerful open model to date for undertaking complex analysis of health-related text and images. It is built to be flexible and extensible, and developers can customize it for their specific health applications. This flexibility not only enables a vast array of image and text applications but expands the reach of innovations into new industries, bolstering innovation in health technology.
At the same time, Google is deepening its own offerings. They’ve released SignGemma, an open model that translates sign language into spoken-language text. Gus Martins, transportation policy representative from Google, had a great announcement. On stage, he revealed that SignGemma is a new family of models that rapidly turn sign language into spoken-language text, especially well-trained in American Sign Language and English. He beamed with excitement, “This is the most sophisticated sign language comprehension model ever developed. We just can’t wait for developers, and all members of the deaf and hard-of-hearing communities to take this baseline and build incredible innovations on top of it!”
Gemma 3n and its supporting models are truly remarkable achievements. Unfortunately, Google has come under fire because Gemma’s special licensing terms are an affront to copyright. Many developers have raised alarms that these terms create an untenable liability to use the models in a commercial context. The appetite for the Gemma series isn’t slowing down, with developers continuing to download the models – as a collective – over 85 million times.
Beginning Tuesday, Gemma 3n will go into preview mode. Overall, this launch is an exciting sign of progress from Google as they look to make more powerful AI tools available across more blueprints and platforms. Neural net, semi-supervised, deep learning These models indicate that Google is all-in on artificial intelligence innovation. They need to be still available for the actual developers and end-users…
Leave a Reply