By default, text-embedding-3-small produces fixed-length embeddings with a predefined dimensionality chosen to balance quality and efficiency. This dimension is consistent across all inputs, which is a critical property for similarity search and indexing. Fixed dimensions ensure that vectors can be compared directly using standard similarity metrics.
From a system design perspective, the default dimension is large enough to encode meaningful semantic information, but small enough to remain efficient in storage and computation. Developers generally do not need to adjust or think about the dimension when starting out. You simply generate embeddings and store them. The consistency simplifies schema design, especially when embeddings are stored alongside metadata in databases or search systems.
This dimensionality works particularly well with vector databases like Milvus and Zilliz Cloud. Milvus requires that all vectors in a collection share the same dimension, and text-embedding-3-small satisfies this requirement naturally. Smaller dimensions also improve indexing speed and query latency. For most applications, the default dimension is a sensible choice that avoids unnecessary complexity while delivering reliable semantic performance.
For more information, click here： https://zilliz.com/ai-models/text-embedding-3-small