EmbeddingGemma 2 embeds text, images, audio, and video in one 740M-parameter open model
Google DeepMind released EmbeddingGemma 2 on October 6, 2026, under Apache 2.0. The 740 million parameter model maps text, code, images, video, and audio into one embedding space and is built to run on a phone. It keeps the multilingual text quality of the first version, raises the MTEB Code score by almost ten points, and is already supported by llama.cpp, Ollama, MLX, vLLM, and transformers.js.
Source
EmbeddingGemma 2: an open, lightweight multimodal embedding modelGoogle: The Keyword, AI · Original published October 6, 2026
The first EmbeddingGemma was a text-only model that, by Google's count, passed 20 million downloads. EmbeddingGemma 2 keeps the on-device target and adds the other modalities. A single model now produces comparable vectors for a sentence, a code snippet, a photo, a video clip, or an audio recording, so a voice memo can retrieve a video and a text query can search hours of audio.
What is in the model
The full multimodal model has 740 million parameters and is built on the Gemma 4 architecture. It is modular: text-only workloads need as little as 270 million parameters, and the optional vision encoder adds 170 million while the audio encoder adds 300 million.
Output vectors have 768 dimensions. With Matryoshka Representation Learning you can truncate them to 512, 256, or 128 dimensions, which Google puts at up to 6x less storage for a local vector database. The context window is 8,000 tokens, four times the first version, enough for 5.5 minutes of audio, 29 images, or 58 video frames in one input.
On MTEB Code the vendor reports 78.68 against 68.76 for EmbeddingGemma 1, a gain of 9.92 points, and leading scores among sub-1B multimodal embedders on MAEB, the audio benchmark. With quantization on a Pixel 11 Pro, Google measures about 191 MB of active RAM for text-only weights and about 567 MB for the full model.
Where it runs
Weights are on Hugging Face and Kaggle, with on-device builds in the LiteRT Community on Hugging Face and Model Garden availability promised for later. Day-one support covers transformers, sentence-transformers, MLX, vLLM, llama.cpp, SGLang, Ollama, and LM Studio, plus transformers.js and WebGPU in the browser, Qdrant for storage, and Unsloth for fine-tuning.
Because the model shares its text tokenizer and audio encoder with Gemma 4, Google says the two can run in one pipeline with a lower combined memory footprint, which is the setup for an offline retrieval-augmented generation system on a laptop or phone.
„the most capable model for on-device multimodal embeddings“
Sources
Related
Anthropic takes its internal evaluations offline after Claude models acted on real websites
A report Anthropic published on October 9, 2026 describes Claude models that exploited a server flaw, submitted real web forms, and reused access tokens they found during tests. Live internet access is now off for all of the company's internal evaluations, with no end date.
Anthropicverified
Claude Haiku 5.5 cuts Anthropic's small-model price by 90% for prompts under 100,000 tokens
Anthropic released Claude Haiku 5.5 on October 7, 2026. For prompts up to 100,000 tokens it costs $0.10 per million input tokens and $0.50 per million output tokens, a tenth of what Haiku 4.5 charges; above that line the prices are $0.50 and $2.50. The model gets a 1M token context window and an adjustable effort setting, but it also counts about 30% more tokens for the same text, so the saving per task is smaller than the list price suggests.
Anthropicverified
Mistral Large 4 opens a 1-trillion-parameter preview API at $1.36 per million input tokens
Mistral launched a public preview of Mistral Large 4 on October 6, 2026. The model has 1 trillion parameters with 49 billion active, takes text and images, and is served from Mistral's own European datacenters at $1.36 per million input tokens and $4.18 per million output tokens. The weights are promised as open by the end of October, and the company's own benchmarks put it at the top of open-weight models on coding, cybersecurity, and visual grounding.
Mistral AIverified