Gemma 3n preview · AI Marketing

Gemma 3n preview

Gemma 3n released today. open source and almost on par with claude sonnet 3.7

available in ai studio now!

This "Google for Developers" blog post announces **Gemma 3n**, a new family of **powerful, efficient, mobile-first open AI models** designed for on-device operation. It's an early preview of technology built on a new, cutting-edge architecture, developed in collaboration with mobile hardware leaders like Qualcomm, MediaTek, and Samsung. This same architecture will power the next generation of **Gemini Nano**, Google's on-device foundation model, set to be available later in the year on platforms like Android and Chrome.

Key highlights of Gemma 3n:

* **On-Device Focus & Efficiency:** Engineered for fast, low-footprint AI experiences directly on phones, tablets, and laptops. It utilizes innovations like **Per-Layer Embeddings (PLE)**, allowing 5B and 8B parameter models to operate with a memory footprint comparable to 2B and 4B models (2GB and 3GB RAM respectively). It boasts ~1.5x faster response on mobile with better quality than Gemma 3 4B.

* **Many-in-1 Flexibility:** A 4B active memory model includes a nested 2B submodel, and a "mix'n'match" capability allows developers to dynamically create submodels to balance performance and quality.

* **Privacy-First & Offline Ready:** Local execution enhances user privacy and allows functionality without an internet connection.

* **Expanded Multimodal Understanding:** Can process audio, text, and images, with enhanced video understanding. Audio capabilities include high-quality Automatic Speech Recognition (transcription) and Translation. It supports interleaved inputs across modalities.

* **Improved Multilingual Capabilities:** Enhanced performance in languages like Japanese, German, Korean, Spanish, and French.

* **Responsible Development:** Built with rigorous safety evaluations, data governance, and alignment with Google's safety policies.

**How to Get Started (Preview):**

Developers can access the Gemma 3n preview via:

1. **Google AI Studio:** For cloud-based exploration (text input).

2. **Google AI Edge:** For on-device development with tools and libraries (text and image understanding/generation).

Gemma 3n aims to democratize access to cutting-edge AI, enabling developers to build new live, interactive, and context-aware applications that process data privately on-device. More details will be shared at Google I/O 2025.

https://developers.googleblog.com/en/introducing-gemma-3n/

0 comments