text-embedding-3-large

Overview. text-embedding-3-large is OpenAI's flagship text embedding model, released in January 2024. It converts text into dense vectors for search, clustering, and retrieval workloads, supporting configurable output dimensions up to 3,072 via Matryoshka Representation Learning, with an 8,191-token input window. Developers reach for it when they need a single, well-supported embedding endpoint for building retrieval-augmented pipelines or semantic search.

Key strengths. The model's dimension-flexibility lets teams trade vector size against storage and search cost without switching models, since a single embedding can be truncated to fewer dimensions while retaining representation quality. Its 8,191-token context window comfortably covers most document chunks, paragraphs, and multi-sentence passages in one call. As a closed-weight, API-only model, it benefits from OpenAI's ongoing maintenance rather than requiring self-hosting.

When to use it. Use this model when you need general-purpose text embeddings for semantic search, retrieval-augmented generation, clustering, deduplication, or recommendation systems, and want a single input-text endpoint rather than a chat interface. It's a good fit when input chunks stay under the 8,191-token limit and when you want the option to store lower-dimensional vectors for cheaper indexing without re-embedding from scratch.

Limitations. This is a non-generative, text-in/vector-out model: it has no chat interface, tool calling, streaming, or system-prompt support, so it cannot be used for generation or reasoning tasks. OpenAI has not disclosed the model's parameter count, architecture, or training-data cutoff, which limits transparency for teams that need that information for evaluation.

Sovereignty note. text-embedding-3-large is built by OpenAI, a US-based creator, so data sent to this endpoint is subject to US CLOUD Act exposure rather than EU-only Sovereign guarantees.