Text Embedding Ada 002
Neural Network
Text Embedding Ada 002 is an OpenAI embedding model for semantic search, RAG, and text clustering.
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 Ada 002?
Text Embedding Ada 002 is a previous generation of OpenAI embedding models: it takes text as input and outputs a numerical vector, used to evaluate semantic similarity between fragments based on proximity. The model is used where an index is already built on it and reindexing is undesirable. In BotHub, it connects via a unified OpenAI-compatible API: access from Russia without VPN or foreign cards, payment in rubles per token, and you'll find over 250 other models in the same window. You can build semantic search over document databases, RAG pipelines over Qdrant or pgvector, support ticket clustering, product card deduplication, and semantic similarity recommendations.Frequently asked questions about Text Embedding Ada 002
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, which can be used to compare the meaning of fragments based on proximity. It is used for semantic search, RAG pipelines, document clustering and deduplication, ticket classification, and recommendations. The context is 4095 tokens, meaning you should pre-split long texts into chunks.
The output is not text or an image, but an array of numbers — a 1536-dimensional vector for each fragment. These vectors are stored in Qdrant, pgvector, Milvus, or another vector database and searched via cosine similarity. The response format is standard JSON from the API.
This is an embedding model; it does not reason or formulate answers — its task is to turn text into a vector. Logic over the found fragments is built by a generative model. In BotHub, you combine them in one window and via a unified API.
In our data, function calling is not marked for this model, and by nature, embeddings do not involve tool calling. However, the API response comes in JSON — it contains the vectors themselves and a token usage counter, which is easy to parse in any backend.
Acceptance of other modalities is not marked in our data: the working input is text strings or an array of them. For images, audio, and video, BotHub has specialized models, and you can switch to them in the same interface or via an OpenAI-compatible API without rewriting the integration.
The model is multilingual and vectorizes Russian texts, so phrases with similar meanings end up close in the vector space. Check the search quality on your own collection: much depends on chunking and the proximity threshold. API access from Russia is available without VPN or foreign cards.