AIToday
Large Language ModelsOpen-Source AIGIGAZINE AIPublished: Oct 7, 2026, 16:00 JST

Google's EmbeddingGemma 2 runs on phones at 191MB RAM

Google's EmbeddingGemma 2 runs on phones at 191MB RAM

3 Key Points

  1. What happened

    Google released EmbeddingGemma 2, its first natively multimodal open embedding model on the Gemma 4 architecture and Apache 2.0 license, with 740 million parameters.

  2. Why it matters

    Because it unifies text, images, video, audio, and code in one embedding space and runs locally on phones and desktops, apps can organize, search, and relate information directly on hardware without an internet connection, enabling on-device search tools and privacy-first RAG.

  3. What to watch

    The small RAM figures are for Google Pixel 11 Pro specifically, so broader device performance hinges on how it runs on other hardware, and the model is described as best-in-class for its 740 million parameter size on benchmarks like MTEB and MAEB.

WHO IT HITSMobile and desktop app developers building on-device search and privacy-focused RAG, who can now embed mixed media locally instead of sending it to a cloud service.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

FAQ
What license is EmbeddingGemma 2 released under?
It is released under the Apache 2.0 license, which Google describes as permissive for commercial use.
What is the model's context window compared to the previous version?
It has an 8K token context window, four times larger than the first-generation EmbeddingGemma.
How much RAM does it need on a phone?
On a Google Pixel 11 Pro, the text-only weights need as little as roughly 191MB of active RAM, while the full multimodal model needs about 567MB.

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleMeta, Walmart, Stripe Back Bret Taylor's Personal Agent Protocol