PINECONE-SERVERLESS

Pinecone Serverless pay-per-use 2024 architecture.

Definition

Pinecone Serverless architecture: (1) Separates storage + compute: embeddings stored S3-like underlying (no dedicated pod), compute provisioned dynamically per query/write request. (2) Pricing: $0.085 per 1M Read Units + $4.00 per 1M Write Units + $0.33 per GB-month storage (us-east-1, slight variations regions), free tier 5M Read Units + 100K Write Units + 5GB storage. (3) Read Units calculation: ~1 RU per query, multiplied by dimension/1000 (1536 dim ~= 1.5 RU per query), pre-filter selective ~+1 RU, large topK >100 ~+1 RU per 100. Sample: 100M queries/month 1536 dim = 150M RU = $12.75. (4) Write Units: similar formula per insert ~1.5 WU per 1536-dim insert. (5) Storage: ~0.4-0.6 GB per 1M 1536-dim vectors with metadata, $0.33/GB-month. (6) Auto-scaling: no infrastructure provisioning + capacity planning, scales 0 to massive instantly. Comparison Pod-Based: Pod-Based fixed compute monthly cost $70+ per s1.x1 pod, Serverless variable based usage typically ~10x cheaper low-volume + comparable high-volume. Migration: Pinecone tools migrate Pod indexes Serverless indexes.

Origin

Pinecone Serverless announced 16 January 2024 ; default offering since ; Pod-Based legacy still available for specific workloads ; pricing public availability all customers immediate.

Example in context

Startup AI app builder migrates Pinecone Pod p1.x1 ($70/month fixed) to Pinecone Serverless: workload ~5M queries/month low-volume = 7.5M RU = $0.64/month + 2GB storage = $0.66/month + 1M Write Units = $4 ; total ~$5.30/month vs $70 = 92% cost reduction.

Related terms

• Pinecone — parent product.