Google’s official Gemini Embedding API documentation explains that embeddings convert text into numerical vectors that can represent semantic relationships between pieces of text. The documentation also notes that developers can specify an embedding task type so the generated vectors better match the intended use case. The listed task types cover semantic similarity, classification, clustering, retrieval documents, retrieval queries, question answering, fact verification, and code retrieval queries. For applications involving search, question answering, classification, or similarity matching, the key editorial takeaway is that embeddings are not presented as a single undifferentiated mechanism; the task type helps shape how the vector representation is intended to be used. This draft is based only on Google’s official documentation and does not add unsupported claims about model availability, pricing, release timing, benchmarks, or product performance. Source: https://ai.google.dev/gemini-api/docs/embeddings