Key Takeaways
- Qdrant, Weaviate, and Chroma are open-source and free to self-host. Pinecone is cloud-only β there is no self-hosted Pinecone at any price.
- Qdrant Cloud has a genuinely free-forever tier (0.5 vCPU / 1GB RAM / 4GB disk, verified 2026-08-27) β no time limit, no card required.
- Pinecone Starter is free (2GB storage, 2M writes/mo, 1M reads/mo); paid tiers start at $20/month flat (Builder) or $50/month minimum usage (Standard).
- Weaviate Cloud has no permanent free tier β only a 14-day Sandbox trial. Paid Flex tier starts at $45/month minimum, verified against Weaviate's October 2025 pricing overhaul, still current as of 2026-08-27.
- Chroma Cloud Starter is $0/month plus usage-based billing (writes, storage, queries); Team tier is $250/month plus usage.
- GitHub stars, verified 2026-08-27 via each project's public API: Qdrant 34,223, Chroma 29,163, Weaviate 16,756.
- Pinecone runs a real public content-creator Affiliate Program (confirmed on pinecone.io/partners) β Weaviate and Qdrant have company-facing partner/reseller programs only, not individual affiliate programs. PromptQuorum is not currently enrolled in any of them.
π In One Sentence
Qdrant wins for self-hosted performance-per-dollar with a free-forever tier; Pinecone wins for zero-infrastructure managed cloud; Weaviate and Chroma fill specific middle-ground niches.
π¬ In Plain Terms
A vector database stores the numerical "fingerprints" (embeddings) of your documents so an AI app can quickly find the most relevant ones β the retrieval half of RAG (retrieval-augmented generation).
Should You Self-Host or Use Managed Cloud?
Self-host if you have DevOps capacity and want to avoid per-request billing; use managed cloud if you want to ship without operating a database. This is the first decision to make, before comparing individual products β it eliminates half the field immediately.
Self-hosting (Qdrant, Weaviate, or Chroma) means you run the database on your own infrastructure. You pay for compute and your own time, not per-query fees, and you control data residency completely. The cost is real: someone has to patch, monitor, back up, and scale that server.
Managed cloud (Pinecone, or Weaviate Cloud / Qdrant Cloud instead of self-hosting) means the vendor runs the infrastructure. You pay a monthly fee plus usage, and uptime, scaling, and patching are the vendor's problem. The cost is a recurring bill that grows with usage, and you depend on the vendor's reliability.
- 1Go self-hosted if: you already run other infrastructure (Kubernetes, Docker Compose stack), need full data residency control, or your vector volume is large enough that usage-based cloud billing would cost more than a VM.
Why it matters: Self-hosting removes per-query billing entirely β a busy RAG app can generate millions of read/write units a month, which adds up fast on usage-based cloud pricing. - 2Go managed cloud if: you are a small team without dedicated DevOps, want to ship in days not weeks, or your vector count is small enough that a free or entry tier covers you.
Why it matters: Managed cloud trades a predictable monthly bill for zero operational burden β no patching, no backup strategy to design, no on-call for a database outage. - 3If unsure, start with self-hosted Qdrant on a small VM or Docker container.
Why it matters: Qdrant's free-forever cloud tier and lightweight Rust binary make it the lowest-risk starting point β you can migrate to Pinecone or Weaviate Cloud later if you decide managed cloud is worth the recurring cost.
Self-Hosted Comparison: Qdrant vs Weaviate vs Chroma
Qdrant, Weaviate, and Chroma are all open-source and free to self-host β Pinecone has no self-hosted option at any price, so it is excluded from this table. All three self-hostable options differ meaningfully on language, license, and deployment complexity.
| Database | Language | License | Hybrid search | Setup difficulty |
|---|---|---|---|---|
| Qdrant | Rust | Apache 2.0 | Yes (built-in) | Low β single binary/Docker |
| Weaviate | Go | BSD-3-Clause | Yes (built-in) | Medium β Docker Compose, modules |
| Chroma | Python/Rust core | Apache 2.0 | Limited (metadata filter + vector) | Lowest β pip install, embedded mode |
πInsight: Chroma is the only one of the three that runs embedded directly inside a Python process with no separate server β the fastest path from zero to a working prototype, but the least suited to a multi-service production deployment.
Pinecone: Managed Cloud, No Self-Hosting
Pinecone is cloud-only β there is no self-hosted version at any price β but its free Starter tier and $20/month Builder tier make it the fastest path from zero to a working managed vector database.** Verified directly on pinecone.io/pricing, 2026-08-27.
Starter (free): up to 2GB storage, 2M writes/month, 1M reads/month, no monthly minimum. Builder: $20/month flat, up to 10GB storage, 5M writes/month, 2M reads/month. Standard: $50/month minimum usage-based billing, unlimited storage at $0.33/GB/month, $4β$4.50 per million write units, $16β$18 per million read units. Enterprise: $500/month minimum, same unit pricing as Standard plus higher-tier support.
Pinecone also offers Bring Your Own Cloud (BYOC) β Pinecone's software runs inside your own cloud account and VPC β which is the closest Pinecone gets to self-hosting, but it is a custom-priced enterprise product, not a free or low-cost self-hosted binary like Qdrant, Weaviate, or Chroma.
Pinecone runs a genuine public Affiliate Program for content creators and educators (confirmed on pinecone.io/partners, separate from its Solution Partner and Referral programs). This page uses plain product links to Pinecone with no current affiliate relationship β PromptQuorum is not currently enrolled in that program.
- Cloud-only β no self-hosted option at any price
- Starter (free): 2GB storage, 2M writes/mo, 1M reads/mo, no minimum
- Builder: $20/month flat, 10GB storage, 5M writes/mo, 2M reads/mo
- Standard: $50/month minimum, unlimited storage, usage-based beyond included volume
- Enterprise: $500/month minimum
- BYOC: custom pricing, runs in your own cloud account/VPC
π‘Tip: Zero-commission check: would we still recommend Pinecone to a startup team choosing managed cloud if Pinecone paid PromptQuorum nothing? Yes β the free Starter tier and $20/month Builder tier are genuinely the lowest-friction way to get a managed vector database running without operating any infrastructure.
Weaviate: Self-Hosted or Managed Cloud
Weaviate is open-source (BSD-3-Clause) and free to self-host, or you can use Weaviate Cloud, which has no permanent free tier β only a 14-day Sandbox trial before you must pay.** Weaviate Cloud restructured its pricing in October 2025: the old $25/month Serverless tier was retired and replaced by a $45/month Flex tier minimum, which is still the current pricing as of this article's 2026-08-27 verification.
Weaviate Cloud Flex: $45/month minimum, shared infrastructure, 99.5% SLA. Plus: $280/month minimum (annual commitment), dedicated infrastructure, 99.9% SLA, SOC 2. Premium: custom pricing, Bring Your Own Cloud, HIPAA-eligible, choice of shared or dedicated deployment. Self-hosted Weaviate has no cost beyond your own infrastructure β it is the same BSD-3-Clause code that runs Weaviate Cloud.
Weaviate's GitHub repository shows 16,756 stars as of 2026-08-27 (verified via GitHub's public API), the smallest of the three open-source options here, though still an actively maintained project with regular releases.
Weaviate has a company-facing partner ecosystem (Technology Alliance Partners, Systems Integrators, Value-Added Resellers) rather than an individual content-creator affiliate program β no per-referral commission structure was found on weaviate.io as of this verification date.
- Self-hosted: free, BSD-3-Clause, unlimited
- Cloud Sandbox: 14-day free trial, not a permanent free tier
- Cloud Flex: $45/month minimum, shared infrastructure, 99.5% SLA
- Cloud Plus: $280/month minimum, dedicated infrastructure, 99.9% SLA, SOC 2
- Cloud Premium: custom pricing, BYOC, HIPAA-eligible
- 16,756 GitHub stars (verified 2026-08-27)
Qdrant: Fastest Self-Hosted, Free Cloud Tier
Qdrant is open-source (Apache 2.0), written in Rust, and has a genuinely free-forever Qdrant Cloud tier (0.5 vCPU / 1GB RAM / 4GB disk, verified on qdrant.tech/pricing 2026-08-27) with no time limit and no card required.** Self-hosting is free at any scale β you pay only for your own infrastructure.
Free tier: 0.5 vCPU, 1GB RAM, 4GB disk, single-node cluster, free cloud inference with selected embedding models. Standard: usage-based pricing beyond the free tier, dedicated resources, flexible scaling, 99.5% uptime SLA β Qdrant does not publish a flat monthly minimum for Standard, unlike Pinecone or Weaviate. Premium: minimum spend required (amount not published), SSO, private VPC links, 99.9% uptime SLA. Hybrid Cloud and Private Cloud: custom pricing for data-residency and regulatory needs.
Qdrant's GitHub repository shows 34,223 stars as of 2026-08-27 (verified via GitHub's public API) β the highest star count of the three open-source options, and Qdrant has raised $50M in funding as of March 2026 to expand its cloud infrastructure.
Qdrant's partner page lists cloud infrastructure partners (AWS, Google Cloud, Microsoft Azure, Oracle Cloud, DigitalOcean, and others for managed deployment) and a startup discount program, not an individual content-creator affiliate program.
- Self-hosted: free, Apache 2.0, unlimited
- Cloud free tier: 0.5 vCPU / 1GB RAM / 4GB disk, free forever, no card required
- Standard: usage-based, no published flat monthly minimum, dedicated resources, 99.5% SLA
- Premium: minimum spend (unpublished amount), SSO, private VPC links, 99.9% SLA
- 34,223 GitHub stars (verified 2026-08-27) β highest of the three open-source options
- Hybrid Cloud / Private Cloud: custom pricing for regulated environments
πInsight: Qdrant is the only one of the four whose free cloud tier has no time limit and no usage-based catch β the 0.5 vCPU/1GB RAM allocation is small, but it never expires and never auto-bills.
Chroma: Embedded, Solo-Developer Friendly
Chroma is open-source (Apache 2.0, Python with a Rust core) and the only one of the four that can run embedded directly inside a Python process with no separate database server at all β the fastest setup path for a solo project or prototype.** Chroma Cloud, its managed offering, is usage-based rather than flat-rate.
Self-hosted/embedded: free, unlimited, no server required for local or single-process use; a client-server mode is also available for shared/production deployments. Chroma Cloud Starter: $0/month plus usage, includes $5 in free credits, up to 10 databases and 10 team members. Chroma Cloud Team: $250/month plus usage, includes $100 in credits, up to 100 databases and 30 team members, adds Slack support and SOC 2 compliance. Enterprise: custom pricing, unlimited databases/members, dedicated support, single-tenant/BYOC.
Usage-based costs on Chroma Cloud, verified 2026-08-27: writes $2.50/GiB, storage $0.33/GiB/month, queries $0.0075/TiB, network egress $0.09/GiB β these apply on top of every tier, including the free Starter tier once the $5 credit is used.
Chroma's GitHub repository shows 29,163 stars as of 2026-08-27 (verified via GitHub's public API), and the project reports usage in over 90,000 other open-source codebases as a dependency.
- Self-hosted/embedded: free, Apache 2.0, no server required for solo use
- Chroma Cloud Starter: $0/month + usage, $5 free credits, 10 databases/10 members
- Chroma Cloud Team: $250/month + usage, $100 credits, 100 databases/30 members
- Enterprise: custom pricing, unlimited scale
- Usage pricing applies on every tier: writes $2.50/GiB, storage $0.33/GiB/mo, queries $0.0075/TiB
- 29,163 GitHub stars (verified 2026-08-27), used as a dependency in 90,000+ repos
Managed Cloud Comparison
Among managed cloud offerings, Qdrant Cloud is the only one with a genuinely permanent free tier β Pinecone Starter is also free but with lower storage limits, and Weaviate Cloud has no free tier at all beyond a 14-day trial.
| Service | Free tier | Entry paid tier | Self-hosted fallback |
|---|---|---|---|
| Pinecone | Starter: 2GB, no minimum | $20/mo Builder flat | None β cloud-only |
| Weaviate Cloud | 14-day Sandbox trial only | $45/mo Flex minimum | Yes β free, BSD-3-Clause |
| Qdrant Cloud | 0.5 vCPU/1GB, forever | Usage-based, no published minimum | Yes β free, Apache 2.0 |
12-Month Cost of 1M Vectors: Self-Hosted vs Managed Cloud
These are illustrative cost estimates, not vendor-published figures β treat the ops-hour and VM-cost assumptions as rough planning inputs, not quotes. Assumptions: 1M vectors at 768 dimensions (~3GB raw), moderate query volume, ops time valued at $75/hour, a small cloud VM for self-hosting at roughly $50β80/month.
Who Should Use Which Database
Match the database to your team's ops capacity and vector volume, not to which one has the most GitHub stars.
- 1Solo/hobbyist project β Chroma embedded
Why it matters: No server to run, no account to create β Chroma runs inside your Python process, which is the fastest path to a working prototype. - 2Privacy-first or regulated workload β self-hosted Qdrant or Weaviate
Why it matters: Full data residency control with no vendor in the data path; both are free, open-source, and production-grade. - 3Small startup shipping fast, no dedicated ops β Weaviate Cloud Flex or Qdrant Cloud
Why it matters: Managed infrastructure without Pinecone's cloud-only lock-in β both give you a self-hosted exit path later if pricing or requirements change. - 4Scaling production app, want zero ops burden β Pinecone Standard or Enterprise
Why it matters: Pinecone's managed-only model is a feature at this scale, not a limitation β no infrastructure decisions left to make. - 5Still not sure β start with self-hosted Qdrant
Why it matters: Free at any scale, fastest engine of the three open-source options, and the free cloud tier gives you a managed fallback with no commitment if self-hosting turns out to be more ops burden than expected.
Skip This Ifβ¦
Skip all four of these dedicated vector databases if your project has under roughly 10,000 vectors and you already run PostgreSQL β the pgvector extension adds vector similarity search directly to a database you are already operating, with no new service to learn, deploy, or pay for. The point where a dedicated vector database starts paying for itself is roughly when query latency or filtering complexity outgrows what pgvector's indexing can handle, not before.
Common Mistakes
Frequently Asked Questions
What is a vector database?
A vector database stores embeddings β numerical representations of text, images, or other data generated by an AI model β and runs similarity search to find the entries most relevant to a query embedding. This retrieval step is the core mechanism behind RAG (retrieval-augmented generation), where a language model is given relevant retrieved context alongside the user's question.
Is Pinecone better than Weaviate?
Neither is better in every situation. Pinecone is cloud-only with no self-hosted option, which is simpler if you want zero infrastructure work but a limitation if you need data residency control. Weaviate offers both a free self-hosted version and a managed cloud option, giving more flexibility, but Weaviate Cloud has no permanent free tier the way Pinecone's Starter does.
How much does it cost to store and query 1 million vectors?
Rough estimates verified 2026-08-27: Pinecone Standard lands near $50β150/month depending on query volume (near the $50/month minimum). Weaviate Cloud Flex stays close to its $45/month minimum at this scale. Self-hosted Qdrant or Weaviate on a small VM runs roughly $50β80/month in infrastructure plus ops labor, which typically makes the 12-month total higher than managed cloud at exactly 1M vectors once labor is counted β see the full TCO comparison above for the reasoning.
How do I migrate from Chroma to Qdrant?
Export vectors and their metadata from Chroma using its Python client (collection.get() with include=["embeddings","metadatas","documents"]), then batch-upload them into a Qdrant collection using Qdrant's Python client's upsert() method, mapping Chroma's metadata fields to Qdrant's payload structure. Both use standard embedding formats, so the vectors themselves need no transformation β only the client-side code and collection schema change.
Is Qdrant the same as Weaviate?
No. Both are open-source, self-hostable vector databases with hybrid search, but Qdrant is written in Rust (Apache 2.0) and Weaviate is written in Go (BSD-3-Clause). Qdrant has more GitHub stars (34,223 vs 16,756, verified 2026-08-27) and a genuinely free-forever cloud tier; Weaviate has a broader module ecosystem for built-in vectorization and reranking but no permanent free cloud tier.
Do Pinecone, Weaviate, Qdrant, or Chroma have affiliate programs?
Pinecone runs a real public content-creator Affiliate Program, confirmed directly on pinecone.io/partners, separate from its company-facing Solution Partner and Referral programs. Weaviate and Qdrant maintain company-facing technology/solution/reseller partner ecosystems rather than individual affiliate programs. No affiliate program was found for Chroma's vector database (trychroma.com). PromptQuorum is not currently enrolled in any of these programs β every link on this page is a plain product link.
Can Chroma be used in production?
Yes, outside of its embedded single-process mode. Chroma's client-server mode and managed Chroma Cloud (Starter from $0 + usage, Team at $250/month + usage) both support multi-service production deployments. Embedded mode specifically β where Chroma runs inside one Python process with no separate server β is best kept to solo projects and prototypes, not shared production workloads.
What is the real difference between self-hosted and serverless/managed pricing?
Self-hosted pricing is infrastructure cost (a VM or server) plus your own ops time for patching, monitoring, and backups β no per-query fees. Managed/serverless pricing (Pinecone, Weaviate Cloud, Qdrant Cloud) charges a monthly fee plus usage-based rates for storage, reads, and writes, with the vendor handling all operations. Self-hosting tends to win on raw dollar cost only once volume is large enough that ops labor is a small fraction of the bill, or when an existing ops team runs the server at near-zero marginal cost.
Which vector database has the most GitHub stars?
As of 2026-08-27, verified directly via GitHub's public API: Qdrant has 34,223 stars, Chroma has 29,163 stars, and Weaviate has 16,756 stars. Pinecone is closed-source and has no public GitHub repository to compare. Star count reflects developer mindshare, not necessarily production suitability for any specific workload.
Do I need a vector database if I only have a few thousand documents?
Probably not. Under roughly 10,000 vectors, the pgvector extension for PostgreSQL handles similarity search well if you already run Postgres, without adding a new service to your stack. Dedicated vector databases like Pinecone, Weaviate, Qdrant, and Chroma earn their operational overhead once vector count, query volume, or filtering complexity outgrows what pgvector's indexing handles efficiently.