Pinecone Adds Gemini Embedding 2 to Documents API Integrated Embedding
- Cargo
- Machine Learning
- Shipped
- Shipper
- William Martinez
- Readers
- 706
Pinecone can embed text for Gemini Embedding 2 during document writes and searches, with plan, index, SDK and multimodal limits.

Pinecone’s Documents API can now use Google’s Gemini Embedding 2 to turn text into vectors during document writes and searches. That brings an additional embedding model into Pinecone’s integrated embedding workflow: teams send text to the database and avoid operating a separate embedding pipeline. The integration is currently limited to text fields on serverless Documents API indexes, so it doesn’t expose the model’s image, video or audio capabilities.

The practical change is in how applications connect the pieces. With an external embedding workflow, an application sends text to an embedding service, receives a vector, then sends that vector to Pinecone for storage or search. With this integration, Pinecone handles embedding as part of those operations. It serves Gemini Embedding 2 through Google Vertex AI and retries failed model calls. Teams don’t need to manage a Google API key or pay Google separately for embedding: token charges appear on the Pinecone invoice at Google’s list price.
Text writes and searches use the same model
To enable the integration, configure a text field’s embed setting when creating a Documents API index. Pinecone then embeds that field’s text on each write and embeds the search query when the application requests semantic search. The field can also have full-text search enabled, letting the same stored text support keyword and semantic retrieval.
For example, an index can embed a support article’s body field at 1,536 dimensions and enable full-text search on it:

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": "gemini-embedding-2",
"dimension": 1536,
},
}
}
},
)After writing documents to the index, an application can submit a text query and ask Pinecone to embed it for question answering:
results = index.documents.search(
namespace="articles",
top_k=5,
score_by=[
{
"type": "embed",
"fields": ["body"],
"query": "how long do refunds take",
"parameters": {"task_type": "question_answering"}
}
],
include_fields=["body"],
)The example selects 1,536 dimensions. Leaving out the dimension setting uses the model’s full vector size; supported dimensions can also accommodate an application’s existing index configuration. Pinecone adds the short input instruction Gemini Embedding 2 expects to distinguish documents from queries, so developers don’t need to manage those model-specific prompts themselves.
Task settings shape the embedding
The default task setting is general search. Developers can instead configure a text field for classification, clustering or sentence_similarity. A search request can set a query-specific task type, including question_answering, fact_verification or code_retrieval_query. These options let a team direct the model toward the job at hand without building a separate embedding call into its application.

Gemini Embedding 2 supports text in more than 100 languages. Google can process requests in any of its regions, but Pinecone doesn’t guarantee that a request will be handled in a particular region. For teams with region-specific processing requirements, that lack of a guaranteed location is a practical deployment constraint.
Availability and current limits
The integration requires a Builder plan or higher and works on serverless Documents API indexes. BYOC, pod-based and Records API indexes aren’t supported. The Pinecone console and Python SDK version 10.3.0 support the model; Node.js and Go SDK support is planned, but isn’t available yet.
Teams with existing vector indexes can convert them to Documents API indexes without re-embedding their vectors. However, adding integrated embedding to an index after conversion isn’t available yet. For now, Gemini Embedding 2 is an option when creating a new Documents API index, where the application can send text directly.
The model’s broader input range also remains out of reach through this integration. Although Gemini Embedding 2 can embed images, video and audio, Pinecone currently offers it only for text fields; multimodal support is planned for a later release. Teams evaluating it for a text retrieval system can try the integrated workflow now, while projects that depend on multimodal embedding will need to wait.
The Builder plan includes 10 million Gemini Embedding 2 tokens per month. Through December 31, 2026, that allowance is doubled to 20 million tokens per month. For teams comparing deployment options, the key trade-offs are straightforward: Pinecone handles embedding and retries in the search workflow, while token charges remain tied to Google’s list price and the available index and SDK support sets the boundaries for adoption.