Google: gemini-embedding-2:free

Gemini Embedding 2 is Google's first natively multimodal embedding model, designed to map text, images, video, audio, and documents (PDFs) into a single, unified vector space. Unlike previous pipelines that required separate models for different modalities, Gemini Embedding 2 understands cross-modal relationships directly. It enables high-accuracy cross-modal retrieval, such as finding a specific video scene using a text description or finding similar audio clips from an image. The model captures deep semantic intent across more than 100 languages and features Matryoshka Representation Learning (MRL), allowing developers to truncate embedding dimensions to save storage and compute costs without significant loss in performance.