StoryFirst seen

Google launches EmbeddingGemma 2 for on-device multimodal embeddings

The 740M-parameter Apache 2.0 model runs multimodal RAG fully offline in about 567MB of RAM on a Pixel 11 Pro.

Archived Updated

Everything it touches

13 names 6 voices 9 source posts

The story’s neighborhood

NAMESVOICESPOSTS

Explore the graph, or choose a connection below.

Follow the evidence

The source posts

This brief comes from Tech Twitter’s edited headlines. The posts below are the sources grouped with the story; statements in them belong to their authors.

Sundar PichaiLead source
Introducing EmbeddingGemma 2, a new open multimodal model that sets the standard for on-device efficiency. - our first open, natively multimodal embedding model - handles text, code, image, video, and audio tasks within a lightweight, modular 740M parameter form factor - ideal for offline, privacy-first RAG when paired with Gemma 4 - outperforms some specialist models more than twice its size Weights available now on Hugging Face.
Read original on X
Google Gemma
Introducing EmbeddingGemma 2! 🚀 Our lightweight, multimodal embedding model maps text, code, images, video, and audio into a single, unified embedding space. Optimized for on-device use cases, it features: - 740M parameter form factor with modular encoders - Flexible dimension sizes (768dim-128dim) via Matryoshka Representation Learning (MRL) - 8K context window (4x larger than text-only EmbeddingGemma) - A commercially permissive Apache 2.0 license
Read original on X
The Information
Today on TITV: Meta and Microsoft are cutting back on employees’ Claude use, and Workday’s AI threat isn’t over. Plus, Reflection’s new open-weight model and a look inside DoorDash’s AI playbook. 📺 Tune in at 10 am PT / 1 pm ET:
Read original on X
Read 5 more source posts
The Information
Today on TITV: Meta and Microsoft are cutting back on employees’ Claude use, and Workday’s AI threat isn’t over. Plus, Reflection’s new open-weight model and a look inside DoorDash’s AI playbook.
Read original on X
Shubham Saboo
What...you can now do multimodal RAG on your phone, fully offline and free. EmbeddingGemma 2 does text search in ~191MB of RAM, or text + image + video + audio in ~567MB. Measured on a Pixel 11 Pro. Open weights under Apache 2.0
Read original on X
etn.
Luke and Ro covered this on the show today: "Meta and Microsoft, two of Anthropic's biggest corporate customers, are working to cut their employees' use of Claude..." "The number of employees using Anthropic's coding assistant Claude Code has dropped sharply by about [50%] recently." "You say to your mum: 'Mum, can we get McDonald's?' and they say, 'No, we have food. We have food at home.' That's basically the AI version." "We've got coding tools at home, guys. We've got AI-enabled coding tools at home." "We've got Copilot in the fridge."
Read original on X
The Information
Exclusive: Microsoft was on track to spend at least $1 billion on Claude for its own employees. It has since cut that projection by more than a third, telling staff to use less Claude and spend more time with Microsoft’s own AI tools. Full story:
Read original on X

How the headline evolved

  1. Google launches EmbeddingGemma 2 for on-device multimodal embeddings

  2. Google launches EmbeddingGemma 2 for on-device multimodal embeddings

  3. Google launches EmbeddingGemma 2 for on-device multimodal embeddings

  4. Google launches EmbeddingGemma 2 for on-device multimodal embeddings

  5. Google launches EmbeddingGemma 2 for on-device multimodal embeddings

  6. Google launches EmbeddingGemma 2 for on-device multimodal embeddings

  7. Google launches EmbeddingGemma 2 for on-device multimodal embeddings

Shared connections

Related stories