Skip to content
TrustList
E2
AI Model

EmbeddingGemma 2

Google's open multimodal embedding model that maps text, images, audio and video into one vector space, under Apache 2.0

About EmbeddingGemma 2

Google released EmbeddingGemma 2 on 6 October 2026 as an embedding model with open weights, published under the Apache 2.0 license on Hugging Face and Kaggle. Availability in Google's enterprise Model Garden is described as coming soon.

One space for five kinds of content

Built on the Gemma 4 architecture, the model maps code, text, images, audio and video into a single shared embedding space. That allows search across formats, so a spoken note can be matched with a video clip, or a typed query with a recording.

Size and structure

The full multimodal model totals 740 million parameters, according to Google. A 270 million parameter core handles text alone, and optional vision (170 million) and audio (300 million) encoders add the other formats. The context is 8K tokens.

Vectors that can be shortened

Output vectors have 768 dimensions, and Matryoshka representation learning allows them to be cut down to 512, 256 or 128. Google says this cuts vector storage by up to six times, which matters for search systems holding very large collections.

Running on a phone

On a phone, the text-only weights need about 191MB of working memory, Google reports. The model therefore fits private retrieval running on the device itself, for material that must stay there.

Cost and comparisons

No hosted rate applies, because users download the model and run it themselves. The benchmark comparisons in Google's announcement are its own, so teams should evaluate retrieval quality on their data before replacing an existing embedding model.

TrustList summary

EmbeddingGemma 2 is Google's open-weight model for turning content into embeddings. Google released it on 6 October 2026 under Apache 2.0, with weights on Hugging Face and Kaggle. It places code, text, images, audio and video in one shared space, where content of one kind can be matched against another, such as a voice memo against a video clip.

What EmbeddingGemma 2 does

  • Shared embedding space: text, image, audio and video content placed in one space where the vectors can be compared.
  • Size and context: 740 million parameters for the full multimodal model, with a 270 million parameter text core and an 8K token context.
  • Optional encoders: vision (170M) and audio (300M) encoders can be added.
  • Adjustable vectors: output can be shortened from 768 dimensions to 512, 256 or 128 through Matryoshka representation learning.

Customers and markets

It suits on-device and private retrieval, where documents need not leave the device. Google puts the active memory for the text-only weights on a phone at about 191MB.

Key facts

  • Released: 6 October 2026
  • Licence: Apache 2.0
  • Architecture: Gemma 4
  • Enterprise access: Google's Model Garden, described as coming soon

Summary written by TrustList from the business's own published information, Nov 2026.

Benchmarks & AI stats

BenchmarkOfficialCommunity avg
No benchmark scores yet. Be the first to add one.

“Official” values are editor-approved and feed the ranking. “Community avg” is the mean of member submissions (shown for transparency; it never affects the ranking until an editor approves a value).

Sign in to add a benchmark score for this model.

EmbeddingGemma 2 launches

What the business tells TrustList

Founded
2026 (0 years) evidence sent

Request a demo or quote from EmbeddingGemma 2

Protected by reCAPTCHA - Google Privacy Policy and Terms apply.

By sending, you agree we may share your request and contact details with the provider once you confirm your email.

Reviews

Write the first review of EmbeddingGemma 2

Used it? Your experience helps other buyers decide.

Write a review

Questions & answers

No questions yet. Be the first to ask about EmbeddingGemma 2.

EmbeddingGemma 2 alternatives

Similar listings buyers compare against this one.

Microsoft AI's high-volume, low-latency variant of MAI-Voice-2.1 for speech in 23 languages, aimed at voice agents

Anthropic's smallest and fastest Claude 5.5 model, available on its own API and three cloud platforms

JetBrains' open 12B mixture-of-experts coding model with 2.5B parameters active per token under Apache 2.0

Open-source 2B decision model from AWS's Strands Agents project that picks options and returns scores with a reliability figure

Microsoft AI's real-time speech-to-text model for voice agents, transcribing 60 languages with continuous language detection

Perplexity's open-weight decision model that picks among given options for routing and checks in AI systems

Ai2's open-weight 8B model that writes cited scientific reports from a research question and retrieved literature

Open-weight translation model from Bilibili's Index team, covering 150 languages and following glossary and style instructions