Model · Google · Embeddings
EmbeddingGemma 2
Google's open multimodal embedding model — puts text, code, images, video and audio into one shared 768-dimension search space. 740M parameters under Apache 2.0, designed by Google to run on laptops and phones, with 100+ languages and an 8K context.
- Modality
- Embeddings
- License
- Apache 2.0 (Open)
- Parameter size
- 740M total: 270M text backbone, 170M vision encoder, 300M audio encoder
- Context window
- 8,192 tokens 8,192 tokens — enough for minutes of audio or video, per Google.
- Released
- October 6, 2026
- Last verified
- October 28, 2026
- Runs locally
- Yes
- Also handles
- vision
What you need to run it
- Mac
- Yes. Any Apple Silicon Mac (M1 or newer) with at least 8 GB (16 GB recommended) of memory. Newer chips are faster, but memory is what decides whether it runs.
- Windows / Linux PC
- NVIDIA works best. A graphics card with at least 4 GB (6 GB recommended) of video memory.
- No graphics card
- Works. Runs on a normal computer's processor, just slower.
Google built it for laptops and phones; MLX, GGUF and ONNX builds are published for Mac.
Numbers are for: 740M model, 16-bit (text-only path is 270M). Our estimate from the model’s size — the maker publishes no figure.
Strengths
- Apache 2.0 — no commercial restrictions
- One vector space for text, images, video and audio, so one index can search across all of them
- Small: the text-only path is 270M parameters, and the vision and audio encoders can be left unloaded
- Matryoshka output: cut vectors from 768 to 512, 256 or 128 dimensions for up to 6x less storage
- 100+ languages, with about 14% better code retrieval than the first EmbeddingGemma
- Works with Sentence Transformers; GGUF and ONNX builds published
Weaknesses
- Needs short task prefixes on text inputs for best quality
- 128-dimension vectors lose noticeable quality outside text-only search
- 8K context is shorter than text-only rivals such as Qwen3-Embedding (32K)
Try it
| Where | Type | Notes |
|---|---|---|
| Hugging Face | weights | Apache 2.0; GGUF and ONNX variants also published |
| Sentence Transformers | local | pip install sentence-transformers |
| Ollama | local | Quantised builds |
Official sources
- Model card model
- Gemma releases docs
Change log
- — Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size and community builds (HF google/embeddinggemma-2 card).
- — Initial entry. Added 2026-10-28 from the popularity sweep: the #1 trending feature-extraction model on Hugging Face, 2,982 likes and 257k downloads/month plus GGUF and ONNX builds. Released 2026-10-06 per Google's Gemma release log (the Hugging Face repo was created 2026-09-14). Apache 2.0, 8,192-token context, 768-dimension output, free to self-host.
Esc