DigestAI news desk

Cut through the AI noise.

Model · embedding model

EmbeddingGemma 2

EmbeddingGemma 2 is an open, lightweight multimodal embedding model with 740 million parameters, built on the Gemma 4 architecture and released under an Apache 2.0 license. Recent news highlights its integration into Google AI Edge Foresight, a Mac-only app that uses local models for meeting transcription and file search without cloud processing.

EmbeddingGemma 2 is an embedding model from Google DeepMind, first covered on 6 October 2026. It is available as open weights you can download and run (licence: Apache 2.0). Its context window is 8K tokens.

Last updated

Quick facts

Who makes EmbeddingGemma 2?
EmbeddingGemma 2 is made by Google DeepMind. Google DeepMind news →
When was EmbeddingGemma 2 released?
The first report of EmbeddingGemma 2's release that Digest AI covered was published on 6 October 2026. EmbeddingGemma 2: an open, lightweight multimodal embedding model →
What type of model is EmbeddingGemma 2?
EmbeddingGemma 2 is an embedding model.
Are EmbeddingGemma 2's weights open?
Yes. EmbeddingGemma 2 is released as open weights that can be downloaded and run, under the Apache 2.0 licence.
What is EmbeddingGemma 2's context window?
EmbeddingGemma 2's context window is 8K tokens.
Where is EmbeddingGemma 2 available?
EmbeddingGemma 2 is available as open weights that can be downloaded and run.

Taken from the coverage, not from our own testing. A question is left out when none of the reports we read gave the answer. All tracked models →

5 stories about EmbeddingGemma 2

  1. Google releases Gemini 4 Argon and announces Gemini 4 Carbon

    Google released Gemini 4 Argon, a frontier reasoning model, to developers, enterprises, and consumers in October 2026. The model is available across Google AI Studio, the Gemini API, Google Cloud, Search, and the…

    2 sources primary source
    Google AI Blognote.com
  2. Research new

    Gemma 4 speech encoder matches larger models on arabic tasks with adapters

    The researchers keep the encoder weights frozen and train a 22.6 M‑parameter CTC reader, then compare it to a version that adds 3.1 M low‑rank adapter parameters, a 41.5 M hybrid CTC/attention reader, and joint…

    1 source primary source
    arXiv cs.CL
  3. Google releases AI Edge Foresight, a Mac note-taking app using on-device Gemma 4 and EmbeddingGemma 2

    Google's AI Edge team released a Mac-exclusive note-taking app called AI Edge Foresight that runs the Gemma 4 and EmbeddingGemma 2 models entirely on device for real-time transcription, summarization, and local…

    1 source
    note.com
  4. Research new

    OpenAI publishes 722 math papers solving 90 of top 500 problems

    OpenAI released 722 mathematical manuscripts from an unreleased internal model, addressing roughly 90 of the top 500 unsolved problems in mathematics. The results include the Quasi-Riemann Hypothesis, described by some…

    1 source
    Latent Space
  5. EmbeddingGemma 2: an open, lightweight multimodal embedding model

    The model builds on the Gemma 4 architecture, contains 740 million parameters and is released under an Apache 2.0 license. It follows the original EmbeddingGemma, which saw more than 20 million downloads for on‑device…

    1 source primary source
    Google DeepMind
Other models from Google DeepMind