## Weaviate against Pinecone, Qdrant and pgvector on cost Of these four, **Weaviate is rarely the cheapest**, and self-hosting it widens the gap, because you also pay for the person who answers the pager when something breaks. ### Free tiers and entry prices Every vendor here offers some way to start small, and none of those entry points tells you much about month twelve. Read each vendor's pricing page on the day you buy, and compare what you would pay at your own data size, not the headline entry price. ### How Weaviate, Pinecone and Qdrant bill compared The four options charge for different things. Weaviate Cloud bills on stored vector dimensions, so a collection with larger embeddings costs more even when the record count stays the same. Pinecone and Qdrant Cloud run as managed services with their own meters and plan floors. pgvector is an extension to Postgres, so you pay for the database server, which you may already be running. ### When pgvector or self-hosting beats Weaviate pricing If your data fits comfortably in a Postgres database you already pay for, pgvector is hard to beat, because you add no new service. Self-hosted Weaviate on Kubernetes **comes to $1,130 for compute** plus $300-600 for storage, so you should expect a bill more than twenty times higher. That gap is the price of running everything yourself, and it only pays off if you need what Weaviate adds beyond plain vector search. ## Bringing down your Weaviate bill without switching To bring the bill down without a new database, make five changes: store smaller vectors, compress them, offload idle tenants, delete stale chunks, and pick the availability tier your workload needs. Because the bill follows stored dimensions, the table sorts each fix by the meter it cuts. Read the last column first, since it tells you which fixes are free and which cost you something. | Fix | Meter it cuts | Expected effect | Trade-off | |---|---|---|---| | Lower-dimension embeddings | Stored dimensions | Largest effect, applied to every object | Re-embed everything; recall can dip | | Compression | Memory per vector | Same data in less space | Slightly less exact matches | | Multi-tenancy offload | Hot storage | Idle customers stop costing full price | Delay when a dormant tenant returns | | Deleting stale chunks | Object count | Stops paying for dead data | Needs a routine to run it | | Availability tier | Replica and floor charges | Drops redundancy you don't use | Less protection during an outage | Whether or not a dimension improves your search, Weaviate bills it. Many embedding models can output shorter vectors, so choose the shortest one that still answers your own test questions correctly. Then turn on Weaviate's documented compression. Check results afterward, because both steps can nudge what comes back. If you serve many customers, each one's data is a tenant. Left alone, a customer who hasn't logged in for months is stored like your busiest account. Offloading inactive tenants moves them out of expensive storage until they're needed, and the cost is a wait when one returns. Doing nothing here is the quiet leak: pages get edited or deleted at the source, and the old chunks stay, billed and still surfacing in answers. A timer-driven workflow fixes that:  1. List what exists in the source. 2. Compare it with what the index holds. 3. Delete chunks whose source changed or vanished. 4. Re-embed only the changed material. Activepieces is an AI automation platform that runs scheduled, deterministic work around a database: triggers, schedules, branches and retries across the apps a team already has, with run history showing what happened. It is a workflow tool beside Weaviate and changes nothing on the Weaviate invoice. The run is the meter, not the steps inside it: the pricing page lists 1 credit per flow run, where Zapier bills per task and Make per module. A ten-step sync with list, compare, delete, embed and upsert each as their own step therefore costs what a two-step shortcut would, so nobody has to cram the work into one fragile step to save money. The Plus plan's 10,000 credits come to $20 a month flat. ## Six steps before committing to a plan Work through six steps in this order, because each one feeds the next and skipping the early ones means you price the wrong thing. 1. Count your chunks, meaning the passages your documents get split into before they are stored. 2. Pick your embedding model and its dimensions, since that choice sets how large each stored integration is. 3. Run both numbers through the pricing page for each plan and each availability tier. 4. Test in the sandbox before it expires. 5. Get support terms in writing. 6. Set a billing alert. ### Count your vectors Every chunk becomes one stored vector, so your chunk count times your dimensions is the real size of your bill. Count after you decide how to split documents, not before. Take both numbers to the pricing page and price each plan, including the minimum floor, which is what you pay even if you store almost nothing. Because the free sandbox expires, load your real data early and run real queries while it's still open. Check that the plan you priced actually handles your workload. Once you're live, set a billing alert so growth in your data shows up as an email, not a surprise. ## What Activepieces does about this The leak this article keeps returning to is stale chunks that stay stored and billed. Activepieces runs the cleanup on a schedule: list the source, compare it with the index, delete what changed or vanished, and re-embed only the new material. The community edition is MIT-licensed and can be self-hosted, so the sync adds no license fee. On the hosted plan, a run is one credit however many steps it has, and Plus is $20 a month for 10,000 credits.  It will not lower the Weaviate invoice by itself. It keeps dead data from growing it. ## Frequently asked questions ### Is Weaviate free to use? You can run the database software yourself for free, but the hosted service isn't free. Weaviate publishes its core code as open source, so you can install it on your own servers without a license fee. You then pay for the machines, storage and the person who keeps it running. For the managed version, Weaviate Cloud, billing follows the plans described earlier. Treat the free software as a trade: you skip the vendor's invoice and take on the operating work yourself. ### Does Weaviate charge per query? Not by the search. Weaviate's cloud pricing is built around what you store and the tier you choose, not around a count of individual searches. That's the pattern on its pricing pages, and it's why the stored vector dimensions matter so much more than how often people search. Check the current plan page before you commit, since billing terms can change. If your traffic is heavy but your data is small, this works in your favor. If your data is large and barely touched, you still pay for it. ### Can I move from Weaviate Cloud to self-hosting later? Yes, because the same open-source database sits underneath both. Weaviate documents ways to back up and restore data, and that's the route most teams would use to move. Plan on real effort: you need to set up servers, security, upgrades and monitoring that the cloud plan handled for you. Doing nothing about that planning is the costly choice, because a move made in a hurry tends to happen during an outage or a billing surprise. ### Does Weaviate offer discounts for annual commitment? For larger or committed customers, Weaviate's pricing pages point toward talking to sales, which is where custom terms are handled. I have no published discount figure to quote, so I won't guess one. If you know your usage will be steady for a year, ask for a quote in writing and compare it against the monthly price of your tier's minimum floor. ### Do I pay separately for the embedding model? Usually, yes. An embedding model is the tool that turns your text or images into the vectors Weaviate stores. Weaviate connects to model providers, and those providers bill you under their own terms. Some Weaviate setups bundle model access, so read the plan details. Either way, count this cost on top of storage, because every document you add has to be converted first. ## Related reading - [Otter.ai Pricing 2026: Free, Pro & Business Plans](https://www.activepieces.com/blog/otterai-pricing-2026-free-pro-business-plans) - [Segment Pricing 2026: Free, Team and Business Plans](https://www.activepieces.com/blog/segment-pricing-2026-free-team-and-business-plans) - [Confluence Pricing: Free and Paid Plans (2026 Guide)](https://www.activepieces.com/blog/confluence-pricing-free-and-paid-plans-2026-guide) ## References - [OpenAPIPricing](https://openapipricing.com/vectordb/weaviate-cloud) - [c17.io](https://c17.io/compare/vector-databases)The software is free to run yourself, but the work isn't.