E5 Large V2
Neural Network
E5 Large V2 is an embedding model that converts texts into 1024-dimensional vectors for semantic search and RAG.
Specifications
- Input
- text
- Output
- vectors (embeddings)
- Context size
- 512 tokens
- Vector dimensions
- 1 024
Prompt cost
(per 1M tokens)
How it works E5 Large V2?
E5 Large V2 maps sentences, paragraphs, and documents into a dense 1024-dimensional vector space, designed for search, ranking, and semantic similarity tasks. The resulting vectors can be directly stored in Qdrant, pgvector, and other vector databases. In BotHub, the model connects via a unified OpenAI-compatible API: one key for all models, access from Russia without VPN or foreign cards, pay-as-you-go in rubles, and encrypted traffic. Typical use cases: knowledge base and documentation search, RAG pipelines for support assistants, clustering of requests and tickets, and re-ranking candidates before responding to users.Frequently asked questions about E5 Large V2
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.
e5-large-v2 by intfloat is an embedding model: it converts text into a numerical vector. Suitable for semantic search, RAG on knowledge bases, document clustering and deduplication, ranking, and classification. Context is 512 tokens, meaning the vector is built from a short fragment or chunk.
Not text or images: the output is an array of floating-point numbers, a vector of fixed dimensionality. The dimension is the same for all requests, so vectors are comparable via cosine similarity and fit into any vector database like Qdrant or pgvector.
Reasoning mode is not indicated in our data, and it is not needed for embeddings: the model does not write an answer or break down tasks into steps; it encodes text into a vector once. You build the search and ranking logic on your application side.
Function calling and structured output are not indicated in our data — these are tools for generative models, while embeddings solve a different task. However, the request itself goes through a unified OpenAI-compatible API, and you receive the response as a standard JSON with an array of vector values.
File input other than text is not indicated in our data, so plan to work with text fragments within the 512-token context limit. If your project requires images, audio, or video, switch to a suitable model in the same BotHub window — you won't need to rewrite the integration.
There is no data on language support in the model card, so we cannot make any promises in advance. Test it on your corpus: index a few hundred documents and evaluate the output quality. Access from Russia is available without VPN or foreign cards, with pay-as-you-go payment for tokens.