Pinecone → Qdrant
What it saves, what actually moves, what you rebuild — and the thing that catches people.
What it costs, and what it saves
Pinecone's serverless pricing scales with storage and reads, and production indexes commonly run $70–500/month. Qdrant self-hosted is the server — a single node handles millions of vectors on modest hardware.
Moves cleanly
The vectors themselves and their metadata, via export and batch upsert.
You rebuild
Client code. The APIs are different shapes, though both are straightforward.
What Pinecone costs you today
Free starter tier (one serverless index, ~2GB). The paid Standard tier carries a $50/month minimum regardless of usage, then usage-based read/write/storage pricing — real workloads commonly land anywhere from a few dollars to hundreds per month.
What actually holds you in
Moderate-to-high. Embeddings themselves are portable (re-upsert them anywhere), but index configuration, namespaces, metadata filters, and any pipeline code written against Pinecone's API need rework. Budget a re-index and an API migration; several alternatives ship Pinecone-compatible import tooling to soften it.
What you are moving to
Qdrant is a high-performance vector database written in Rust and licensed Apache-2.0 — free to self-host with no usage limits or feature gates. It ships as a single self-contained binary with REST and gRPC APIs, runs via Docker or Kubernetes (official Helm chart and Operator), and offers a managed Qdrant Cloud on AWS, GCP, and Azure when you'd rather not operate it. Strong filtering, quantization, and hybrid search make it the most common 'we left Pinecone' landing spot.
Free self-hosted (no limits); Qdrant Cloud managed tiers with a free 1GB cluster
Qdrant strengths
- Apache-2.0, no feature gates — the full engine is open
- Single Rust binary: laptop to cluster with the same API
- Excellent metadata filtering, quantization, hybrid search
- Managed cloud exists when you want zero ops
What you give up
- Self-hosting means you own scaling and backups
- Smaller managed-service ecosystem than Pinecone's
The migration, step by step
- 1Export vectors from Pinecone in batches — there is no bulk dump, so paginate through
- 2Deploy Qdrant with Docker and create a collection with matching dimensions and distance metric
- 3Batch upsert, including payloads; Qdrant's payload filtering is richer than Pinecone's metadata filters
- 4Swap the client library and rewrite query calls
- 5Compare recall on a held-out query set before cutting over
The gotcha
Distance metric must match exactly. Indexing with cosine when the embeddings were built for dot product produces results that look plausible and are subtly wrong — the worst kind of bug because nothing errors.
When to stay on Pinecone
You want zero operational burden and the bill is not yet painful.
More Layer 4 migrations
Figures verified 2026-07-15 against vendor pricing pages. Prices change and migrations differ by estate — treat the cost delta as a starting model, not a quote. Rankings and recommendations here are merit-only; affiliate income never changes a verdict. See our methodology.