Google releases EmbeddingGemma 2 paper: unified multimodal embeddings beating models 12x larger
osanseviero · x · 2026-11-03
The EmbeddingGemma team at Google published the EmbeddingGemma 2 paper (arXiv:2610.37364), an open multimodal embedding model:
- Unified space: natively represents text, images, audio, video, and combinations in a single embedding space.
- Method: simplified per-layer embeddings plus a multi-stage curriculum with modality/objective progression.
- Results: establishes a new Pareto frontier across all-modality embedding benchmarks, outperforming specialist models and models 2–12x larger.
- Efficiency: selective modality encoder loading, adjustable vision token budgets, truncatable embedding dimensions, and quantized weights for resource-constrained settings.
Includes ablations of key design decisions; the model is released openly for further research.
Related event: Google Open-Sources EmbeddingGemma 2 for On-Device Multimodal Search(2 posts)→
More from Models
- Rumor: OpenAI model solved Navier-Stokes millennium problem and 300+ open math problems at ~3h compute each — haider1 · 2026-11-03
- AgentTime Update: GPT Models Lead Time-Awareness, Best Forecast Miss Is 2.26× — maksym_andr · 2026-11-03
- paul_cal catalogs the untold failures of LLM tuning — paul_cal · 2026-11-03
- DeepSeek V5 reportedly to be pretrained on Huawei Ascend chips, Ascend 960DT due Q1 2027 — teortaxesTex · 2026-11-03
- Rumors swirl of massive GPT-6.5 pretraining run reportedly using millions of GPUs — VraserX · 2026-11-03
- Tsinghua spinoff Apex Intelligence open-weights Apex-35B, built by automated AI research system — jiqizhixin · 2026-11-03