Pinecone Turns Gemini Embeddings Into a One-Step Search Engine

Pinecone is putting Google’s Gemini Embedding 2 directly inside its managed document search workflow. Announced on November 12, 2026, the integration lets applications send plain text to Pinecone while the Documents API creates embeddings for both writes and search queries through Vertex AI.

That removes a layer of infrastructure that developers previously had to assemble themselves. There is no separate Google Cloud account, API key, or embedding pipeline to manage, and embedding usage appears on the Pinecone invoice instead of being billed separately by Google. The managed option costs $0.20 per 1 million tokens.

One API for Text, Vectors, and Search

Embeddings are vectors that represent meaning, allowing software to compare text by what it means rather than only by the exact words it contains. With Gemini Embedding 2 inside the Documents API, the same field can support similarity-based retrieval and keyword search without the application generating vectors first.

The workflow is designed to keep document content and its embedding together. Pinecone writes each document’s text and embedding at the same time, so semantic search and full-text search use the same version of that document. Pinecone invokes the model on every write and query, then retries failed calls when needed.

Customers choose gemini-embedding-2 in a text field’s embed configuration when creating an index. Pinecone’s model documentation lists 3,072 dimensions by default, while any integer from 128 to 3,072 is supported. Cosine is the sole similarity metric, and each input can contain up to 8,192 tokens.

Those choices are locked when the index is created: the model, dimension, and metric cannot change afterward. The first release supports serverless Documents API indexes, but excludes BYOC, pod-based, and Records API indexes.

Task Types Add Control Without Prompt Engineering

Gemini Embedding 2 uses task types to handle different workloads. General retrieval uses retrieval_document for stored text and retrieval_query for searches by default, helping the model treat the document side and query side for their different jobs.

Classification, clustering, and sentence-similarity tasks treat stored text and queries alike for text comparisons. Other query settings cover question answering, fact verification, and code retrieval, giving applications more ways to shape how the model represents their data.

A single search can override the field’s default query task. Pinecone inserts Gemini’s required task instruction automatically, so users submit plain text instead of constructing model-specific prompts. The integrated-embedding documentation also fixes a field’s write task when the index is created.

Those hidden instructions add about six or seven tokens to each value or query, and those tokens count toward the model’s input limit. Charges apply to embedding tokens on both writes and queries, making the cost model important for applications that update large document collections or run frequent searches.

• Embedding requests can contain up to 96 inputs.
• Each search query is limited to 10,000 bytes.
• The model requires a Builder, Standard, or Enterprise plan.
• The model is unavailable on Starter.
• The Builder plan normally includes 10 million Gemini Embedding 2 tokens per month.
• Pinecone doubles that allowance to 20 million through December 31, 2026.

What Developers Can Use Today

Pinecone lists console support and Python SDK 10.3.0 support for the announcement. Node.js and Go SDK support is coming soon, giving teams more ways to connect applications to the managed workflow.

Existing vector indexes can be converted to Documents API indexes without re-embedding. Adding integrated embedding to a converted index is not yet available, so the current workflow uses new Documents API indexes when teams want Pinecone to create embeddings for them.

Multimodal support is planned for a later release. For now, the integration centers on text, but it already changes where the operational work lives: developers send content, Pinecone handles model calls and retries, and one invoice tracks the embedding usage.

That shift matters because search systems often fail at the seams between document storage, embedding generation, model configuration, and billing. Gemini Embedding 2 does not erase those design decisions, since teams still choose dimensions, task types, metrics, plans, and index settings. It does give them one managed path for putting those decisions into production.

With the first release focused on serverless Documents API indexes, Pinecone is establishing the foundation for a broader search workflow. The next question is how far the integration will expand when Node.js and Go support arrive, converted indexes gain integrated embedding, and the planned multimodal release moves beyond text.

Stay connected via Google News

Follow us for the latest AI News