Start free, scale effortlessly

Pod-based indexes let you decide up front how much storage capacity and compute resources to use. Choose pod-based indexes if you’re certain about your workload and require throughput of more than 5 Queries Per Second (QPS) on a single namespace. Recommender systems typically fit this description. Pod-based indexes are billed hourly based on the number and type of pods used.

Serverless indexes let you start and scale with ultimate ease and flexibility, without provisioning resources. They use our next-generation architecture, algorithms, and API to provide the best developer experience, scalability, and cost efficiency, especially for larger indexes with lower throughputs per namespace. Multi-tenant RAG and labeling/classification workloads typically fit this description. Serverless indexes are billed for the usage of reads, writes, and storage.

Usage information in API responses and in your dashboard will help you monitor and plan for costs. View documentation on how usage is calculated , then start testing with your own data and workloads. Contact us for help estimating costs for large or complex workloads.

Example 1: Multi-tenant RAG workload with 5M total documents across 100,000 users (namespaces)

(1536 dim embeddings)

Example 2: Semantic search for enterprise knowledgebase with 10M vectors:

(1536 dim embeddings)

Example 3: Offline data labeling using similarity search over 1B vectors

(1536 dim embeddings)

Pinecone serverless will soon be generally available in AWS regions, followed by public-preview availability for Azure and GCP regions.

Pod-based indexes are generally available across multiple regions in AWS, Azure, and GCP.

Pinecone serverless is currently available on AWS regions ‘us-east-1’, ‘us-west-2’, and ‘eu-west-1’. Support for additional regions is coming soon.