Developers have applied the model to build DigUp, an open-source tool for Mac that indexes local files and podcasts. To facilitate on-device search, Google uses Matryoshka embeddings to store only the first 256 of 768 values in a vector. This approach reduces storage needs to 1/3 while maintaining about 95% of the retrieval quality for visuals and speech, according to Google.
Tuesday, October 20, 2026
Story
Google's EmbeddingGemma 2 Rivals OpenAI on Consumer CPUs
AI7 days ago
The newly released 740M parameter multimodal model from Google has quickly become the 4th most liked in the company's history after surpassing EmbeddingGemma 1 in popularity. EmbeddingGemma 2 benchmarks comparably to OpenAI's embedding-3-large and Qwen 8B while running on consumer CPUs rather than GPUs. Though it successfully identified 13 of 14 video clips from a single sentence in testing, it lacks a sense of temporal direction, viewing a video and its reversed version as 99% similar.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.