Google EmbeddingGemma 2
Product Hunt Rohan Chaubey ● Covered by 7 sources
Google launched EmbeddingGemma 2, an open model for text, code, images, video, and audio. It can search across all of them from one query, and it runs on-device.
Based on reporting by Product Hunt, Rohan Chaubey — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Google DeepMind has launched EmbeddingGemma 2, an open embedding model built to turn text, code, images, video, and audio into one shared space. The pitch is simple: one model, one query, and search that can reach across very different kinds of content without forcing everything into the same old text box.
The model comes in at 740M parameters and is meant to run on-device. Google says the full multimodal version needs about 567MB of RAM on a Pixel 11 Pro, while the text-only version needs 191MB. That keeps it in the realm of software that can live closer to the user, rather than sitting far away in a server and waiting its turn.
There’s also a clear quality story here. On MTEB Code, Google says the score moves from 68.76 to 78.68. That is not a tiny nudge; it is the kind of jump that makes a model look much more serious for code retrieval work, which is exactly where embedding models earn their keep.
The release is Apache 2.0 licensed, with weights available on Hugging Face and Kaggle. So this is not a walled garden debut. It is an open model being pushed out with practical distribution channels, and that makes it easier for developers to try it, compare it, and build with it without first signing up for a sermon about the future of AI.
My take — AI-written commentary, not fact-checked reporting
Open models win when they’re useful, not when they come with a victory speech, and EmbeddingGemma 2 at least shows up with numbers. The real tell is the on-device angle: privacy and latency are the boring parts that actually matter, which is why they keep getting hand-waved by louder product launches.
Read more about this at: Product Hunt