Pinecone adds OpenAI search embeddings without a separate account or API key

Pinecone customers can now send plain text for search without first calling OpenAI themselves. On November 10, 2026, Pinecone added text-embedding-3-small and text-embedding-3-large to its integrated embedding service. The change moves model calls and billing into Pinecone, with no separate OpenAI account or API key, but initially requires new serverless Documents API indexes.

The same models, less work between services

With a separately managed setup, developers call OpenAI’s embedding API for every document write and search query. They also manage the OpenAI account, credentials and rate limits, then pay OpenAI separately from Pinecone. The new option moves those responsibilities into the database service rather than introducing a new OpenAI model.

Developers select an embedding model for a text field. Pinecone then embeds each write and query for that field; a search request can contain the query as plain text. Pinecone also writes a document’s text and embedding together. The company says this keeps full-text and semantic search working from the same document version.

The hosting arrangement also shifts recovery work. Pinecone says it serves the models through Microsoft Azure across several regions. If one region slows or returns errors, the service retries elsewhere within the same processing location and redirects traffic until the affected region recovers. Teams do not have to build that failover themselves.

Existing indexes cannot switch on the feature yet

The immediate adoption limit is the index type. These models work with serverless Documents API indexes, not bring-your-own-cloud deployments, pod-based indexes or Records API indexes. Pinecone’s availability details therefore narrow the launch to one route through its service, rather than every existing Pinecone deployment.

Existing vector indexes can be converted to Documents API indexes without generating their embeddings again. However, adding integrated embedding to a converted index is still coming soon. Until then, Pinecone directs customers to create a new Documents API index and send it their text. Conversion alone does not unlock this integration.

• Available now: the Pinecone console and Python SDK 10.3.0 support both models.
• Coming soon: support in the Node.js and Go SDKs.
• Processing choices: global processing is the default; US-only processing requires the Standard plan or above. Both models are available on every plan.

Model settings and a temporary token allowance

Setup happens when creating the index: developers put the model in a text field’s embed setting. Pinecone’s example assigns text-embedding-3-small to a body field and also enables full-text search on that field. The same stored text can then serve semantic searches and keyword searches, rather than requiring separate text fields for each.

Developers can also choose the embedding’s dimensions, or representation size. The small model supports 1 to 1,536 dimensions, while the large model supports 1 to 3,072. Omitting the setting selects each model’s full size. Pinecone supports any size within those ranges, including dimensions already used by a customer’s existing indexes.

Embedding tokens are billed on the Pinecone invoice at OpenAI’s list price. Starter normally includes 5 million tokens monthly for each OpenAI model, and Builder includes 10 million per model. Through December 31, 2026, Pinecone is doubling those allowances to 10 million and 20 million respectively. The increased allowance is temporary, not a permanent change to either plan.