Gemini Embedding 001
Neural Network
Gemini Embedding 001 is a Google vector representation model for semantic search, RAG, and text clustering.
Specifications
- Input
- text
- Output
- vectors (embeddings)
- Context size
- 20 000 tokens
- Vector dimensions
- 3 072
- Capabilities
- Batch API
Prompt cost
(per 1M tokens)
How it works Gemini Embedding 001?
Gemini Embedding 001 is a Google embedding model that converts text into vectors and holds top positions in the MTEB benchmark. It is designed to work across multiple domains: science, law, finance, and programming code. In BotHub, the model is available via a unified OpenAI-compatible API, without VPN or foreign cards, with payment in rubles based on usage. Requests are transmitted encrypted, and data is not stored. Vectors can be stored in Qdrant or pgvector to build semantic search over document databases, RAG pipelines for support assistants, ticket clustering, product card deduplication, and finding similar code snippets in repositories.Frequently asked questions about Gemini Embedding 001
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
The model converts text into numerical vectors: these are used to build semantic search, RAG pipelines, clustering and deduplication, ticket classification, and recommendations. A context of up to 20,000 tokens allows vectorizing large document fragments entirely, without splitting them into small pieces, and the BATCH mode is convenient for mass database indexing.
The output is not text, but an array of numbers — a fixed-length vector for each fragment. The response arrives in JSON via a unified OpenAI-compatible API and can be directly stored in vector databases like Qdrant or pgvector, without intermediate format conversion.
No separate reasoning mode is noted in our data, and the model's task is different: it does not write answers, but returns vectors. Build reasoning logic on a text model to which you pass the fragments found via embeddings — switching it in the same API is easy.
There are no notes about function calling or structured output schemas in our data. However, the embedding response already arrives as a structured JSON with an array of vectors, so a separate mode is usually not required. If you need tools and strict schemas, connect a text model using the same key.
Acceptance of anything other than text is not noted in our data, so plan your pipeline around text fragments: transcripts, recognized speech, descriptions, and file captions. For working with images, audio, and video, use specialized BotHub models — integration and access remain the same.
Gemini family models are designed for multilingual corpora, but specific language characteristics are not in our data, so evaluate the quality on your own documents: collect a set of typical queries and check the relevance of the output. Access to the model from Russia is available without VPN or foreign cards, with payment in rubles.