EmbeddingGemma 2: an open, lightweight multimodal embedding model
- Source
- blog.google
- Date

Context
Google DeepMind released EmbeddingGemma 2 on October 6, 2026, an open model that turns text, code, images, audio and video into vectors, lists of numbers in one shared space, so a text query can be compared by similarity with a video clip or voice memo. Google's announcement says it has 740 million parameters, uses the Apache 2.0 license and is designed to run on-device, where search and retrieval can work offline and keep data local.
Google reports MTEB Code, a code-retrieval benchmark, rising from 68.76 for the first EmbeddingGemma to 78.68. Its claims of leading scores among sub-1B models come from Google's own evaluations, which it refers to a model card.
The size is adjustable: text alone needs about 270 million parameters, optional vision and audio encoders add the other modalities, and vectors can be truncated from 768 to 128 dimensions for up to six times less storage.
- The model is modular: text-only workloads need as little as 270M parameters, with optional vision (170M) and audio (300M) encoders added for full multimodal support.
- Google says that with on a Pixel 11 Pro it needs about 191MB of active RAM for text-only weights and about 567MB for the full multimodal model.
- Matryoshka Representation Learning lets developers truncate vectors from 768 dimensions to 512, 256, or 128, cutting local vector storage and memory by up to 6x.
- The is 8K tokens, four times the first EmbeddingGemma, enough for about 5.5 minutes of audio, 29 images, or 58 video frames in one input.
- Code retrieval improved most: MTEB Code rose from 68.76 to 78.68. Because it shares Gemma 4's text tokenizer and audio encoder, running both together in one pipeline lowers combined memory.
- multimodal — A model that works with more than text — reading images, audio, or video, and sometimes generating them too.
- embedding — A list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.
- quantization — Shrinking a model by storing its numbers less precisely — like rounding — so it runs faster and fits on smaller hardware, at a small quality cost.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
Checking sign-in…
Loading comments…

