Pinecone Adds OpenAI Text Embedding 3 Models to Integrated Search
- Cargo
- Machine Learning
- Shipped
- Shipper
- Thomas Davis
- Readers
- 914
Pinecone can now embed documents and queries with OpenAI’s text embedding 3 models inside Documents API indexes, with usage billed through Pinecone.

Pinecone’s integrated embedding now supports OpenAI’s text-embedding-3-small and text-embedding-3-large, bringing two widely requested models into its Documents API workflow. Teams can send text directly to Pinecone for indexing and search, without running a separate embedding service in their application.

The practical change is in the operational path: Pinecone handles embedding on document writes and search queries, while storing text and vectors together. That can simplify systems already built around Pinecone, especially where developers would otherwise maintain provider credentials, make embedding calls, and coordinate separate billing. The models join the Pinecone-hosted embedding options that were available when integrated embedding launched.
Usage for OpenAI’s models appears on the Pinecone invoice at OpenAI’s list price. No OpenAI account or API key is needed for this integration. Starter and Builder plans receive double their normal monthly included token allowance for these models through December 31, 2026: 10 million tokens per model each month on Starter and 20 million per model each month on Builder. The standard allowances are 5 million and 10 million, respectively.
One path for text, vectors, and search
With a conventional setup, an application sends text to an embedding provider, receives a vector, and then sends that vector to the search system. Queries need the same treatment. Integrated embedding lets an application submit plain text to Pinecone, which generates the embedding and performs the search as part of the same request.

For writes, Pinecone stores a document’s text and its generated embedding together. Keeping those representations in the same record means full text and semantic search use the same stored version of a document. That reduces the coordination work involved in updating two separate systems when source text changes.
Pinecone says it serves the OpenAI models through Microsoft Azure across several regions. If a region slows down or returns errors, Pinecone retries in another region within the same processing location and shifts traffic away from the affected region until it recovers. This takes regional retries and traffic handling out of the application’s embedding path, though the service still depends on the availability of the configured processing location.
In the Python SDK, developers select a model in the embed setting for a string field when creating a Documents API index. For example, a support knowledge base can embed its article body with text-embedding-3-small at 1,536 dimensions and enable full text search on that same field. Documents are then upserted as text, without a separate vector generation step.

from pinecone import Pinecone
pc = Pinecone(api_key="YOUR_API_KEY")
pc.indexes.create(
name="support-articles",
schema={
"fields": {
"body": {
"type": "string",
"full_text_search": {},
"embed": {
"model": "text-embedding-3-small",
"dimension": 1536,
},
}
}
},
)
index = pc.index("support-articles")
index.documents.upsert(
namespace="articles",
documents=[
{"_id": "kb-205", "body": "Refunds take five business days."},
],
)Search requests can also use plain text. The query is associated with the field to embed, and the request can return the stored body alongside its match results.
results = index.documents.search(
namespace="articles",
top_k=5,
score_by=[
{
"type": "embed",
"fields": ["body"],
"query": "how long do refunds take",
}
],
include_fields=["body"],
)Dimension choice gives teams room to match an existing vector size or use the model’s full output. text-embedding-3-small supports dimensions from 1 to 1,536, with 1,536 as its default. text-embedding-3-large supports 1 to 3,072 dimensions, defaulting to 3,072. If the dimension setting is omitted, Pinecone uses the model’s full size.
Availability and constraints
The models are available on every plan, but US-only processing requires the Standard plan or above. By default, text is processed globally; selecting the US processing location limits processing to the United States. That location choice is a deployment consideration for teams with specific data handling requirements.
Support is limited to serverless indexes using the Documents API. BYOC, pod-based, and Records API indexes aren’t supported. The Pinecone console and Python SDK version 10.3.0 support the models; Node.js and Go SDK support is planned, but not yet available.
Existing vector indexes can be converted to Documents API indexes without re-embedding their data. However, adding integrated embedding to a converted index is not yet supported. For now, teams can use the OpenAI models on new Documents API indexes and send their text through the integrated workflow.
The setup is most useful when a team wants Pinecone to own both embedding and search requests, and accepts the supported index types, processing locations, and per-model dimension ranges. The remaining adoption details to watch are SDK support beyond Python and whether integrated embedding becomes available on converted indexes.
Setup instructions are available in the integrated embedding documentation, with index creation details in the Documents API guide.