Text Embedding 3 Small
Neural Network
Text Embedding 3 Small is an OpenAI embedding model that converts text into vectors for semantic search and RAG systems.
Specifications
- Input
- text
- Output
- vectors (embeddings)
- Context size
- 8 192 tokens
- Vector dimensions
- 1 536
Prompt cost
(per 1M tokens)
How it works Text Embedding 3 Small?
Text Embedding 3 Small is an OpenAI vector model, an evolution of the ada line: it encodes text into numerical vectors used to measure the semantic similarity of fragments. Comparison is based on vector distance rather than word matching, making the model a foundational layer for search, classification, and ranking. In BotHub, calls are made via a unified OpenAI-compatible API: access from Russia without VPN or foreign cards, pay-as-you-go in rubles, with over 250 models available in one window. It is used for semantic search across document databases, RAG for corporate wikis, deduplication and clustering of requests, finding similar materials, and populating vector stores like Qdrant or pgvector.Frequently asked questions about Text Embedding 3 Small
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.
This is an embedding model from OpenAI: it turns text into a numerical vector, which is convenient for calculating semantic similarity. It is suitable for semantic search, RAG pipelines, clustering and classification of documents, deduplication, and recommendations. It accepts up to 4095 tokens as input, so long texts are split into fragments.
The output is not text, but a fixed-dimension vector — an array of floating-point numbers. It is stored in vector databases like Qdrant or pgvector and searched using cosine similarity. Requests are made via the unified OpenAI-compatible BotHub API.
There is no indication of reasoning before answering in our data. The embedder's task is different: it encodes the meaning of text into a vector, rather than unfolding a chain of thought. If you need reasoning and explanations, connect a text model via the same API — you won't need to rewrite the integration.
Function calling and structured output are not noted in our data. The embedding model's response is a vector, which your application places into the required structure itself. For tools and JSON schemas, it is more convenient to use a BotHub text model, accessing it via the same key.
The acceptance of modalities other than text is not noted in our data: you pass text fragments and receive vectors. If your project requires images, audio, or video, specialized models are available in BotHub — they can be connected alongside in the same interface and API.
The text-embedding-3 family from OpenAI was trained on multilingual data, so Russian texts are vectorized correctly and cross-lingual search works. Accuracy still depends on your domain, so run a test sample and compare metrics on your documents before going to production.