Pinecone is a managed vector database — the part of a RAG or semantic-search stack that stores embeddings and finds the nearest matches fast. Here's how its serverless pricing works, what the free Starter tier gives you, and how to get a key.
| Plan | Price | What you get |
|---|---|---|
| Starter free tier | $0 | serverless, ~2 GB storage, good for testing |
| Standard | from ~$25/mo min | usage-based: storage + reads + writes, pay-as-you-go |
| Enterprise | from ~$500/mo min | higher limits, SSO, SLA, support |
The Starter plan is free and serverless — roughly 2 GB of storage, no credit card to start, which is plenty for a few hundred thousand embeddings and a working RAG demo. The catch with vector DBs is that cost scales with both how much you store and how often you query, so a chatbot that's cheap at launch can creep up as the knowledge base and traffic grow. Model it before you commit.
1. Sign up at pinecone.io.
2. Open the console and create an index (pick the dimension to match your embedding model, e.g. 1536 for OpenAI text-embedding-3-small).
3. Go to API Keys and copy your key.
4. Connect with the SDK using that key and your index host.
Upsert and query (Python):
If you already run Postgres, pgvector is free and often enough below a few million vectors. Qdrant and Weaviate are open-source and self-hostable; Chroma is great for local prototyping. Managed rivals like Milvus/Zilliz compete on price at scale. Pinecone wins on being fully managed and low-ops. Pair it with embeddings + an LLM from OpenAI or Claude, and feed it pages scraped with Firecrawl. To estimate the whole stack, use our AI cost calculator.
Yes — the Starter plan is free and serverless with about 2 GB of storage and no credit card, enough to build and test a small RAG app.
Sign up at pinecone.io, create an index in the console (matching your embedding dimension), then go to API Keys and copy the key. Use it with the SDK plus your index host.
Serverless billing is usage-based: you pay for storage (per GB-month) plus reads and writes (per million units), with a monthly minimum on paid plans. Cost grows with both data size and query volume, not per user.
pgvector is free if you already run Postgres and is usually cheaper up to a few million vectors. Pinecone costs more but is fully managed, scales further and needs no ops work — the trade-off is money for time.
Not affiliated with Pinecone. Prices are reference estimates — always verify on the official pricing page.