Overview

EmbeddingGemma is derived from Gemma 3 and trained for over 100 languages. With quantization it runs in under 200 MB of RAM, which makes private, on-device retrieval possible.

Highlights

  • 308M parameters
  • 100+ languages
  • Matryoshka embeddings (768 down to 128 dimensions)
  • 2K-token inputs

Typical uses

On-device semantic search, RAG and classification.

Things to consider

  • Gemma terms of use apply.
  • Download sizes refer to the default quantized versions in the Ollama library; full-precision weights are larger.

Specifications are taken from the model card on Hugging Face, the Ollama library and the provider’s announcement. Benchmark results depend on evaluation settings; see how AI benchmarks work.