What happened
Pinecone's announcement, published November 10, 2026, says its integrated embedding feature now supports OpenAI's text-embedding-3-small and text-embedding-3-large alongside Pinecone-hosted models. A team configures either model on a string field when creating a Documents API index, then sends plain text for document writes and search queries. Pinecone generates the embedding for each write and query, and the application can search through the Documents API rather than calling an embedding provider first and managing those calls itself.
The announcement describes the operational change as moving embedding work into Pinecone. Text and its embedding are written together, so full-text and semantic search use the same version of a document. Pinecone says it serves the OpenAI models through Microsoft Azure across multiple Azure regions. If a region slows down or returns errors, Pinecone retries in another region in the same processing location and shifts traffic away until recovery. The company says this removes the need for teams to build their own failover for these embedding calls.
The models support configurable dimensions. text-embedding-3-small accepts dimensions from 1 to 1,536, with 1,536 as its default; text-embedding-3-large accepts 1 to 3,072, with 3,072 as its default. Pinecone says any supported size can be used, which allows teams to keep dimensions already used by existing indexes. Processing is global by default, or limited to the United States when the field uses the US processing location. The US processing option is available on Standard plans and above.
Pinecone says the models are available on every plan but work only on serverless Documents API indexes; BYOC, pod-based, and Records API indexes are not supported. The console and Python SDK 10.3.0 support the models. Node.js and Go SDK support is described as coming soon. Existing vector indexes can be converted to Documents API indexes without re-embedding, but adding integrated embedding to a converted index is also described as coming soon. Until that capability arrives, Pinecone directs users to new Documents API indexes.
Tokens are billed on the Pinecone invoice at OpenAI's list price, rather than on a separate OpenAI bill. Starter normally includes 5 million tokens per month for each OpenAI model and Builder 10 million. Through December 31, 2026, those amounts are doubled: 10 million per model per month on Starter and 20 million on Builder. Pinecone's example uses text-embedding-3-small at 1,536 dimensions and enables full-text search on the same body field, illustrating a single field for keyword and semantic retrieval.
What the archive adds
New Runtime's earlier coverage on October 23, 2026, described Pinecone adding integrated embedding to the Documents API with three launch models and combined semantic and full-text search. That earlier note makes this a new fact on a known story: the Documents API now adds two OpenAI embedding options, with their own limits and billing details.
Trend signal
This extends Pinecone's effort to make search functionality available within an application-facing product boundary, but it does not establish a broad new agent capability or protocol. This does not yet fit a pattern New Runtime tracks.
Why now
Pinecone says the OpenAI models are among its most requested embedding options. The doubled Starter and Builder token allowances run through December 31, 2026, giving eligible teams a time-bounded reason to test the integration.
Why it matters
For a team already using a supported Documents API index, integrated embedding can reduce the number of services and credentials involved in retrieval. The application can submit text, while Pinecone handles embedding and search; combining semantic and full-text retrieval on one field also reduces the chance that a document's keyword-search text and vector represent different versions.
That convenience comes with a boundary. Teams on BYOC, pod-based, or Records API indexes cannot use this release as described, and teams requiring US-only processing need Standard or above. The token price remains OpenAI's list price, so the announcement does not claim cheaper embeddings; the change is primarily operational simplicity, consolidated billing, and Pinecone-managed regional failover. Teams should compare that trade-off with their existing provider controls and index architecture before migrating.
Limits and uncertainty
Pinecone's announcement describes availability and integration behavior but does not report independent measurements of latency, retrieval quality, reliability, or total operating cost compared with a separately managed OpenAI pipeline. It also does not quantify the Azure regional failover performance. Node.js and Go support and integrated embedding for converted indexes are future capabilities, without dates in the announcement. The token allowance increase is explicitly temporary, ending December 31, 2026.
Skill to build now
Test the integration against a small slice of a real corpus this week. Configure one text field with the embedding model and dimensions your current index uses, enable full-text search if needed, then compare representative queries against your existing retrieval path. Include both semantic and exact-term queries, and verify the billing and processing-location settings before moving production traffic.
Reader takeaway
Pinecone now handles OpenAI text-embedding-3 calls inside supported Documents API indexes, consolidating the workflow without lowering the stated token price. Check index type, processing location, and monthly token limits first: the integration is not yet available for BYOC, pod-based, or Records API indexes.