Introducing EmbeddingGemma: our new open, state-of-the-art embedding model designed for on-device AI 📱

EmbeddingGemma can work together with models like Gemma 3n to power advanced generative AI experiences and RAG pipelines.

The model is the highest ranking open multilingual text embedding model under 500M on the MTEB leaderboard, is offline-first by design, and works with your favorite open source tools.

The small, fast, and efficient model leverages QAT to reduce RAM usage, features a 2K token context window, is trained across 100+ languages, and allows devs to build powerful apps that run directly on-device.

Start building with llama.cpp, @lmstudio, and @ollama. Download the models on @huggingface and @kaggle:

@ollama @huggingface @kaggle

EmbeddingGemma + Hopper == low latencies and massive throughput. Try it for yourself on Baseten:

Google devs are the most consistent ones...dropping something every other day.. another SOTA

Congrats 👏 Google is crushing it

Perfect timing for a project I am on. ty

Don’t sleep on the embedding models

Now available on Msty app as well. Get the best local RAG with the best local embedding model 😎

Slow and steady wins the AI race

Straight model link for those who can't wait:

On-device AI is empowering. This opens doors to personalized experiences while preserving privacy.

amazing！more and more，i am looking forward to Apple🤗

🔥🔥 Btw is there any open source something like Google's multimodalembedding model which takes image -> embedding as well as text -> embedding in same dimensional vectors from same model, a tiny one.

EmbeddingGemma sounds solid for devs. Using it offline-first is a game changer for mobile apps. What's the community feedback on performance so far?

Could it run in an iPhone?

I use Gemini now, but looking for Gemma especially for my Real Estate maintenance work. Cool to see your applied case!. we mostly operate in applying AI to help people with service. Would love to connect! Please DM.

impressive

Yesss another thursday of model releases

how th did your read my mind?!

@grok O que é isso? Pra que serve e pra quem é interessante?

On-device? That’s great!

@grok what's the usecase of this Gemma since we already got gemini 2.5 flash and 2.5 pro

@grok para que serve isso e compare com por exemplo o cursor que está no mercado

Exciting to see Google pushing the boundaries of on-device AI.

@grok eu consigo fazer download e utilizar o modelo offline? A licença dele MIT , ou seja , eu posso utilizar na minha empresa?

Google is flexing 💪🏽 geeks rejoice 🦾

Excited to see this in action.

Do we have to create a ‘UI’ for this?

@grok j’aimerais que tu m’expliques, à quoi ça sert ce type de modèle concrètement ?!