Embedding Models in Model Library

What's New

Embedding models are now integrated into the Model Library — browse, filter, and use them directly when configuring assistants for retrieval, search, and RAG workflows.

Available Models

• OpenAI: text-embedding-3-large (3072d), text-embedding-3-small (1536d), ada-002 (1536d).

• Google: gemini-embedding-001 in 768, 1536, and 3072 dimensions.

• Cohere: embed-v4.0 (256–1536d), embed-english-v3.0, embed-multilingual-v3.0, embed-multilingual-light-v3.0.

New API Endpoints

• GET /v1/embeddings — List all available embedding models.

• GET /v1/embeddings?provider=openai — Filter by provider.

• GET /v1/embeddings?dimensions=3072 — Filter by vector dimensions.

• GET /v1/embeddings/:id — Get model details including pricing and capabilities.

How to Use It

Specify embedding_provider and embedding_model_name directly when creating or updating assistants. The system automatically handles chunking, dimension matching, and vector store indexing.

Migration Notes

• Fully backward compatible — existing integrations unchanged.

• Default embedding model remains text-embedding-3-small unless explicitly changed.

• Switching embedding models on an existing assistant triggers automatic re-indexing of stored documents.