Google DeepMind · 06 Oct 2026
EmbeddingGemma 2
Google’s open, lightweight on-device model for multimodal embeddings across text, images, audio and video.
Material model update, access expansion, pricing shift or comparable capability change.
Release is backed by the lab’s own announcement or newsroom post.
Verified source · 1 receiptModel lineage
Where EmbeddingGemma 2 sits.
Model lifecycle
Know whether it is ready—and when it is leaving.
Radar tracks the model’s current access stage separately from its original announcement, including official retirement dates and recommended replacements when the lab publishes them.
Model weights are published under the stated licence.
- Status since
- 06 Oct 2026
Change Lens
What actually changed.
Normalized from the release note so different labs can be compared by the same fields, not just by launch copy.
Multimodal · Open weights
Evidence is first-party; comparison labels are normalized by Radar.
Release is backed by the lab’s own announcement or newsroom post.
Verified source · 1 receiptSource receipts
Evidence behind the brief.
The Radar keeps claims traceable: first-party sources, normalized fields, unknowns and the latest verification marker used for this page.
Pricing, Context are not clearly specified in the indexed source note.
Maps multiple modalities into one embedding space while remaining compact enough for private, on-device retrieval and classification.
This brief links back to the first-party announcement or documentation used by the radar.
Open original source