At a glance
| Product type | Proprietary managed vector database (serverless); no self-hosted server edition |
| Vector index types | Not user-selectable: dense, sparse and document indexes; pod-based indexes are legacy |
| Hybrid search | Yes: dense plus sparse or BM25 full-text fields, with optional reranking |
| Metadata filtering | Yes: flat key-value metadata, all fields indexed by default, 40 KB per record |
| Embedding integration | Pinecone-hosted embedding and reranking models, or bring your own vectors |
| Scaling and multitenancy | Serverless scaling; namespaces per tenant; dedicated read nodes on paid plans |
| Deployment (managed or self-hosted) | Managed on AWS, Google Cloud and Azure; BYOC for Enterprise |
| Licensing and pricing model | Closed source; free Starter, $20 Builder, usage-based Standard ($50 min) and Enterprise ($500 min) |
Pinecone suits teams that want vector search without running any database infrastructure and are comfortable with a closed source service. The free Starter plan and $20 Builder plan cover prototypes; production use on Standard starts at a $50 monthly minimum plus read, write and storage units, so model your query volume before committing.
Overview
Pinecone is a managed vector database run by Pinecone Systems. You create indexes through its API or console, upsert records that hold vectors and metadata, and query them by similarity. There is no open source or downloadable server edition: the database runs on Pinecone's serverless platform on AWS, Google Cloud or Azure, or, for Enterprise customers, in their own cloud account through BYOC (generally available since 25 August 2026).
Pinecone does not expose a choice of ANN algorithm such as HNSW or IVF; you choose a record model instead. Document indexes (generally available on 2 September 2026 under API version 2026-07) combine dense vector, sparse vector and full-text string fields in one schema, while legacy vector indexes hold either dense or sparse vectors. Pinecone also hosts embedding and reranking models, so text can be embedded at upsert and query time without a separate model service.
Key features
- Serverless indexes billed on read units, write units and storage, with no capacity to size up front
- Document indexes that mix dense vectors, sparse vectors and BM25 full-text fields in one schema (GA September 2026)
- Namespaces that partition an index per tenant, from 100 per index on Starter up to 1,000,000 on Enterprise
- Metadata filtering with operators such as $eq, $in, $gt, $and, $or and $not, up to 40 KB of metadata per record
- Integrated inference: Pinecone-hosted embedding and reranking models called at upsert and query time
- Dedicated read nodes for large, high-throughput indexes on Standard and Enterprise (GA April 2026)
- Pinecone Assistant, a managed document question-answering service billed separately by tokens and storage
Integrations and ecosystem
Pros and cons
Pros
- No servers, shards or index parameters to manage on serverless indexes
- Published unit prices and plan minimums, plus a permanent free Starter plan
- Dense, sparse and full-text search in one document index, with hosted embedding models
- Namespaces give a simple per-tenant partitioning model at large tenant counts
Cons
- Closed source with no self-hosted edition outside Enterprise BYOC, which raises lock-in and data residency questions
- Read unit billing makes costs depend on query volume and result size, which is harder to forecast than fixed capacity
- Starter is limited to AWS us-east-1, 5 indexes and 2 GB of storage
- Pod-based indexes are being phased out, so older deployments need a migration to serverless
Pricing
Pinecone publishes five plans. Starter is free: 2 GB storage, 2M write units and 1M read units a month, up to 5 indexes, 1 project and 2 users, on AWS us-east-1 only. Builder is a flat $20 a month with higher limits and no usage overages. Standard has a $50 monthly minimum and Enterprise a $500 monthly minimum, both billed on usage: read units from $16 to $18 per million on Standard ($24 to $27 on Enterprise), write units from $4 to $4.50 per million ($6 to $6.75 on Enterprise), storage $0.33 per GB per month, and egress $0.10 per GB after 100 GB a month, with rates varying by cloud and region. Bring Your Own Cloud (BYOC) is priced by sales. Standard has a 3-week trial with $300 of credits. Prices are shown with a $ sign; the page does not name the currency code.
Pricing checked on the vendor's official page on 7 November 2026. Prices change; confirm before buying.
Who it is best for
Application teams building retrieval-augmented generation or semantic search who want a fully managed vector store with hosted embeddings and are happy with a proprietary service.
Typical use cases
- Retrieval-augmented generation over company documents
- Semantic and hybrid product or support-article search
- Per-customer vector stores in a multi-tenant SaaS product using namespaces
- Recommendation and similar-item lookup from embeddings
Who should look elsewhere
- Teams that need an open source database they can run anywhere: compare Qdrant, Weaviate or Milvus.
- Projects whose vectors belong next to relational data in PostgreSQL, where pgvector avoids a second system.
Verdict
Pinecone is the most hands-off option in this category: serverless indexes, hosted embedding models and published unit prices make it quick to start. The trade-offs are a closed source service with no self-hosted edition for most buyers and a usage-based bill that grows with query volume, so teams that need portability or predictable fixed costs should compare open source options.
Frequently asked questions
Is Pinecone open source?
No. Pinecone is a proprietary managed service. Its SDKs are public, but the database itself cannot be downloaded; Enterprise customers can run it in their own cloud account through BYOC.
Does Pinecone have a free plan?
Yes. Starter is free with 2 GB storage, 2M write units and 1M read units a month, up to 5 indexes, on AWS us-east-1. Builder costs a flat $20 a month for higher limits.
Can I still create pod-based indexes?
Generally no. Pinecone treats pods as legacy: new Standard and Enterprise organisations since 18 August 2025 cannot create them, and SDKs using API version 2026-07 no longer create them. Existing pod indexes keep working and can be migrated to serverless.
Does Pinecone support hybrid search?
Yes. You can combine dense vectors with sparse vectors, and since September 2026 document indexes also support BM25 full-text fields alongside vector fields in one index.
Sources
- Pinecone pricing
- Pinecone indexing overview (documentation)
- Pinecone 2026 release notes
- Understanding pod-based indexes (legacy)
- Pinecone pods vs serverless migration guide
- Pinecone integrations
Checked November 2026.
Listed in: Vector Databases & AI Search.
How we research reviews and what star ratings mean: our editorial method.