Embed 5 Fast
Neural Network
Embed 5 Fast is a multimodal Cohere embedding model for semantic search and RAG.
Specifications
- Input
- text
- Output
- vectors (embeddings)
- Context size
- 128 000 tokens
- Vector dimensions
- 2 048
Prompt cost
(per 1M tokens)
How it works Embed 5 Fast?
Embed 5 Fast is the lower-latency and lower-cost variant of Cohere's Embed 5 family: a multimodal model with approximately 1 billion parameters that converts text, images, and mixed fragments into a common vector space. In BotHub, the model is available via a single OpenAI-compatible API: access without VPN or foreign cards, payment in rubles based on usage, over 250 models in one window, AES-GCM encryption without data storage, and for companies — contracts, invoices, and electronic document management. Scenarios: semantic search in documentation, RAG over knowledge bases, clustering and deduplication of requests, multimodal image search, and vector export to Qdrant and pgvector.Frequently asked questions about Embed 5 Fast
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.
embed-5-fast is an embedding model: it converts text into a vector used for semantic search, RAG, and clustering. It is convenient for knowledge base search, finding similar documents, and deduplication. The context is up to 128,000 tokens, so long texts fit into a single fragment.
The model does not generate text, images, or files — its result is a vector representation of the input text, which you save in a vector database. The response format is a numerical vector: it is convenient to store in Qdrant, pgvector, or another vector DB. Context is 128,000 tokens.
Reasoning before answering is not observed in embed-5-fast, which is expected: the model does not conduct dialogues or provide explanations. Its task is to immediately return a vector for the input text. Reasoning chains and step-by-step explanations should be requested from text models, and the result then indexed by this embedder.
Function calling and structured JSON output are not observed. This is not a chat model: it takes text as input and provides a vector representation as output, not a JSON message. If you need function calling and strict JSON, it is better to assign such tasks to a text model and use embeddings for search.
Acceptance of images, audio, and video is not observed in embed-5-fast: the model works with text, and the vector is built from a text fragment. Images and recordings must first be converted into text or descriptions, then sent for vectorization. Check the documentation for details on supported input types.
We do not have specific data on the quality of performance in Russian, so we cannot promise anything specific. It is easy to check: take your 'query-document' pairs, build vectors, and see if the search finds what is needed. For mixed collections, test each language separately.