Pinecone adds OpenAI’s text-embedding-3 models to integrated embedding
Pinecone now supports OpenAI’s text-embedding-3-small and text-embedding-3-large in its integrated embedding service, allowing teams to use the models for semantic search without maintaining a separate embedding pipeline.
- Pinecone added text-embedding-3-small and text-embedding-3-large.
- Users can send plain text for writes and queries.
- Tokens are billed on the Pinecone invoice at OpenAI’s list price.
- Starter and Builder token allowances are doubled through December 31, 2026.
- The models are available on serverless Documents API indexes.
Embedding and search in Pinecone
The two OpenAI models join the Pinecone-hosted models available through integrated embedding. Teams can select either model for a text field in a Documents API index and send Pinecone plain text. Pinecone then creates embeddings for every write and query associated with that field.
This removes several steps from a typical setup. Instead of calling OpenAI before sending data to Pinecone, a development team can send text directly. Pinecone embeds a query and runs the search in one call, while writing each document’s text and its embedding together. According to Pinecone, this helps full-text and semantic search use the same version of each document.
No separate OpenAI account or embedding bill
With a direct OpenAI integration, teams need to manage an OpenAI account, API key and rate limits, and pay OpenAI separately from Pinecone. Pinecone’s integrated option moves those embedding calls into its service. There is no OpenAI API key to manage and no separate bill for embedding. The tokens appear on the Pinecone invoice at OpenAI’s list price.
Pinecone says it serves the models from Microsoft Azure, with traffic distributed across several Azure regions. If a region slows down or returns errors, Pinecone automatically retries in another region within the same processing location and shifts traffic away until the region recovers.
Models, dimensions and plan limits
Both models support configurable dimensions. text-embedding-3-small supports dimensions from 1 to 1,536, with 1,536 as its default. text-embedding-3-large supports dimensions from 1 to 3,072, with 3,072 as its default. Pinecone says users can leave out the dimension to use the model’s full size, or select a supported size that matches an existing index.
The models process text globally by default, or only in the United States when a field uses the US processing location. They are available on every plan, while the US processing location is available on the Standard plan and above.
Through December 31, 2026, Starter and Builder plans receive twice their included monthly tokens for OpenAI’s models. Starter receives 10M tokens a month on each model, and Builder receives 20M a month on each model. The regular allowances are 5M tokens for each OpenAI model on Starter and 10M on Builder.
Availability for Documents API users
The models work on serverless Documents API indexes. They are not supported on BYOC, pod-based or Records API indexes. The Pinecone console and Python SDK 10.3.0 support them, while support in the Node.js and Go SDKs is coming soon.
Pinecone’s example creates a Documents API index with a body field using text-embedding-3-small at 1,536 dimensions and enables full-text search on that field. Users can then upsert plain text and search with a text query, allowing the same field to support semantic and keyword search.
Existing vector indexes can be converted to Documents API indexes without re-embedding, but adding integrated embedding to a converted index is coming soon. For now, Pinecone directs users to new Documents API indexes for these models.
