Google releases EmbeddingGemma 2, an open-source multimodal embedding model
What happened
On 7 October 2026, Google released EmbeddingGemma 2, an open-source multimodal embedding model that converts text, images, video, audio and code into numerical vectors, with 740 million parameters. Google says it beats competitors with twice the parameter count on multimodal embedding benchmarks, scoring 78.68 on the Massive Text Embedding Benchmark (Code), nearly 10 points above the previous generation's 68.76. The same day, Google DeepMind released the model, saying it is based on the Gemma 4 architecture and uses the Apache 2.0 licence, mapping text, images, audio and video into a unified embedding space. Earlier reports said the model could handle text, images, video, audio and code, while the later DeepMind release notes listed text, images, audio and video as the modalities. The same day, Simon Willison commented on the Apache 2.0 licence that embedding models should not be closed-source, proprietary, hosted-only, because users typically need to compute and store thousands or even millions of embedding vectors, and if a vendor stops offering the model, users have to pay to recompute those stored vectors; he would rather pay for a hosted version while keeping the option to run open weights himself or find another supplier if the vendor shuts down. On 8 October, Google released Google AI Edge Foresight, an experimental Mac app that fully offline transcribes meetings and audio files, using the 740-million-parameter on-device EmbeddingGemma 2 model to record meeting notes across apps, including in-person meetings, and is free to use.
Written by AI from the reports · updated 8 d ago
Developments
- Oct 8 21:28 · 2 reportsGoogle releases local-first AI note-taking app ForesightTechCrunch · AI：Google releases a new local-first Granola competitor
- Oct 7 03:47 · 3 reportsGoogle releases EmbeddingGemma 2 embedding modelGoogle DeepMind：EmbeddingGemma 2: an open, lightweight multimodal embedding model
Timeline
Follow the reports to see the event from each side.
- The Verge · AIGoogle’s AI note-taking app transcribes your meetings completely offline
Google has released an experimental note-taking app that can transcribe meetings and audio files entirely offline, as reported earlier by TechCrunch. The app, called Google AI Edge Foresight, is free to use and runs on macOS using the company's on-device EmbeddingGemma 2 model.
- TechCrunch · AIGoogle releases a new local-first Granola competitor
Google’s new AI Edge Foresight app takes on Granola with an offline meeting note-taker that can transcribe conversations, generate notes, and answer questions using on-device AI.
- Simon WillisonEmbeddingGemma 2
My comment on EmbeddingGemma 2 — Hacker News. I really appreciate that EmbeddingGemma 2 is under the Apache 2.0 license. For embedding models in particular, I don't think it makes sense to use a closed, proprietary, hosted-only model. Most applications of embedding models involve calculating thousands or even millions of embedding vectors and storing them for later comparison.
- Google DeepMindSelectedEmbeddingGemma 2: an open, lightweight multimodal embedding model
EmbeddingGemma 2 is the most capable model for on-device multimodal embeddings, natively mapping combinations of text, images, audio, and video into a unified embedding space.
- The DecoderGoogle claims EmbeddingGemma 2 outperforms rival embedding models twice its size
Google released EmbeddingGemma 2, an open model with 740 million parameters that converts text, images, video, audio, and code into vectors. It runs on-device, needs only about 191 MB of RAM, and outperforms some competing models twice its size, according to Google.
Heat of this event
Current heat 3·Peak in the comparable range 12（Oct 9 16:00）·Change over 24 hours in the comparable range -75%
The trend compares only the same accounts observed without a break, so its range can be narrower than the current heat count. Move the pointer or tap the chart to see each hour; with a keyboard, use the left and right arrow keys.