Google’s Gemini Embedding 2 Now Available: A Deep Dive

Google has officially released Gemini Embedding 2, its first natively multimodal embedding model, through the Gemini API and Vertex AI. This advancement marks a significant step in AI’s ability to understand and process diverse data types, offering developers powerful modern tools for a wide range of applications.

What is Gemini Embedding 2?

Gemini Embedding 2 expands upon previous text-only foundation models by mapping text, images, videos, audio, and documents into a unified embedding space. This means the model can understand the semantic intent across over 100 languages, simplifying complex AI pipelines and enhancing tasks like Retrieval-Augmented Generation (RAG), semantic search, sentiment analysis, and data clustering. [Google Blog]

Key Capabilities and Specifications

• Text: Supports input contexts of up to 8192 tokens.
• Images: Processes up to 6 images per request, accepting PNG and JPEG formats.
• Videos: Handles up to 120 seconds of video input in MP4 and MOV formats.
• Audio: Natively ingests and embeds audio data without requiring intermediate text transcriptions.
• Documents: Directly embeds PDFs up to 6 pages long.

The model’s ability to understand interleaved input – combining multiple modalities like image and text in a single request – allows it to capture nuanced relationships within complex, real-world data. [Google Blog]

How Gemini Embeddings Work

Gemini Embedding 2 utilizes dense vector representations, similar to those used in large language models. Unlike sparse vectors that directly map words to numbers, dense vectors represent the meaning of text. This allows for more accurate searches based on semantic alignment, even if the query and passages don’t share the same keywords. [Google Cloud Documentation] The vectors are normalized, enabling the use of cosine similarity, dot product, or Euclidean distance for similarity rankings.

Availability and Pricing

As of August 31, 2025, the Gemini Embedding model is generally available through the Gemini API and Vertex AI. [ByteTrending] It is priced at $0.15 per 1M input tokens and has a maximum input token length of 2048. [ByteTrending]

Applications of Gemini Embeddings

Gemini Embeddings are particularly well-suited for:

• Retrieval-Augmented Generation (RAG): Enhancing the accuracy and relevance of generated content.
• Semantic Search: Finding information based on meaning rather than keywords.
• Clustering: Grouping similar documents or data points together.
• Sentiment Analysis: Determining the emotional tone of text.

Gemini Embedding vs. Other Models

The Gemini Embedding model has consistently ranked #1 on the MTEB (Multilingual Translation Evaluation Benchmark) Multilingual leaderboard since its experimental launch in March, demonstrating exceptional performance across a wide range of languages. [ByteTrending]

Vertex AI also offers other text embedding models, such as gemini-embedding-001, which uses 3072-dimensional vectors. [Google Cloud Documentation]