Vector Databases for AI Agents Compared (2026 Pricing)

Every agent that remembers anything needs somewhere to keep it. Past conversations, retrieved documents, tool results and user preferences usually end up as embeddings in a vector database, and the agent queries that store on almost every turn. That makes the database choice a cost and latency decision, not just an engineering one. This guide compares five managed options developers pick most often for agents: Pinecone, turbopuffer, Qdrant Cloud, Weaviate Cloud and Chroma Cloud.

Prices as of Nov 4, 2026, taken from each vendor's official pricing page (linked in Sources). Vendors change rates often, so check the page before you commit.

Why agents stress a vector database differently

A classic RAG app does one search per user question. An agent can do several per step: recall memory, search docs, check earlier tool output, then write new memories back. Three things follow.

• Writes matter. Agents write constantly (new memories, summaries, tool results), so write pricing and minimum write sizes show up on the bill.
• Multi-tenancy matters. Most agent products keep one memory space per user or per project, which means thousands of small namespaces or tenants.
• Hybrid search matters. Agents often look up exact names, IDs or error strings, where keyword (full-text) search beats pure vector similarity.

Pricing at a glance

Each vendor bills on different units, so a direct per-query comparison is not possible from list prices alone. Here is what each one charges for, in its own units.

• Pinecone: free Starter (up to 2 GB storage, 2M write units and 1M read units a month). Builder is $20 a month flat. Standard has a $50 monthly minimum, then $0.33 per GB per month storage, $4 to $4.50 per million write units and $16 to $18 per million read units, depending on cloud and region. Enterprise has a $500 monthly minimum.
• turbopuffer: no free tier; the Launch plan has a $16 monthly minimum. Usage is billed on logical bytes: storage up to $0.33 per GB, writes up to $2.00 per GB, and queries billed on data scanned (up to $1.00 per PB). Enterprise starts at $4,096 a month.
• Qdrant Cloud: a free single-node cluster (0.5 vCPU, 1 GB RAM, 4 GB disk). Standard is billed hourly on the resources you provision (vCPU, memory, storage and backups) with a 99.5% uptime SLA. Premium needs a minimum spend.
• Weaviate Cloud: free tier with 100,000 objects, 1 GB memory and 10 GB disk. Flex is pay-as-you-go with a $45 monthly minimum, billed on vector dimensions (from $0.00465 per million) and storage (from $0.12 per GiB). Premium is a prepaid contract that starts from $400 a month.
• Chroma Cloud: Starter is usage-based with $5 in free credits; writes cost $2.50 per GiB, storage $0.33 per GiB per month, queries $0.0075 per TiB queried and $0.09 per GiB returned. Team is $250 a month plus usage, with $100 of credits.

Pinecone: the safe default

Pinecone is fully managed and serverless, with dense, sparse and full-text indexes on every plan. The new $20 Builder plan fills the gap between the free tier and the $50 Standard minimum, which helps solo developers who outgrow 2 GB. For agents, the namespace limits are the number to watch: 100 per index on Starter, 1,000 on Builder and 100,000 on Standard. If you plan one namespace per user, that pushes you to Standard early.

Pinecone also sells embeddings, reranking and a managed Assistant, so one bill can cover the whole retrieval stack. See the Pinecone tool page for more.

turbopuffer: cheapest at scale for many small namespaces

turbopuffer stores data in object storage and caches what you query, which is why storage is cheap. Its calculator example, 1 million 1,024-dimension documents with 1 million writes and 1 million queries, comes to about $7 of usage, below the $16 Launch minimum. That model suits agents with lots of per-user memory that is rarely all queried at once.

Two billing details matter. Every write is billed at a minimum of 10 KB, so many tiny memory writes cost more than you expect; batching earns up to a 50% discount at around 3.2 MB per batch. And queries are billed on at least 1.28 GB of data per namespace, even for small namespaces. See the turbopuffer tool page.

Qdrant Cloud: pay for the cluster, not the call

Qdrant prices by provisioned resources, billed hourly, rather than per read or write. That makes costs predictable for steady, high-query agent workloads, because a busy hour costs the same as a quiet one on the same cluster. The trade-off is that you size the cluster yourself, and idle capacity still costs money. Qdrant is open source, so you can also self-host it or use Hybrid Cloud to run managed clusters in your own infrastructure. See the Qdrant tool page.

Weaviate Cloud: built-in agent features

Weaviate bills on the total number of vector dimensions stored plus storage, and compression lowers the dimension rate. Its free tier is generous for prototypes but limited to one collection and three tenants. The notable extra for agent builders is the Query Agent, which turns natural-language questions into database operations: 1,000 free requests a month, or $30 per organization monthly with 4,000 requests included. Weaviate updated its price book on September 1, 2026, changing backup rates and adding RQ-4 compression pricing.

Chroma Cloud: the open-source favorite, now serverless

Chroma is the database many developers first meet in local agent prototypes. Chroma Cloud keeps the same API and adds serverless vector, full-text and metadata search. Its calculator example, 1 million documents written, 6 million stored and 10 million queries a month at 1,536 dimensions, comes to about $79. Pricing is simple, but the Team plan jumps to $250 a month once you need SOC 2 and Slack support.

Which one should your agent use?

• Prototype or side project: Chroma (local, then Cloud credits), the Qdrant free cluster or the Weaviate free tier.
• One memory space per user, at scale: turbopuffer, or Pinecone Standard if you want more built-in services.
• Steady, high query volume: Qdrant Cloud, where you pay for capacity rather than every call.
• Natural-language querying out of the box: Weaviate with the Query Agent.
• Strict data residency: Pinecone BYOC, Qdrant Hybrid Cloud or self-hosted open-source Qdrant, Weaviate or Chroma.

If your agent runs on AWS, our guide to deploying an AI agent on AWS covers the compute side. For model access, see our comparisons of AI gateways and Claude on the API, Bedrock and Vertex AI.

Pros and cons

• Pinecone: pro, the most complete managed stack; con, the $50 Standard minimum and per-unit pricing that is hard to estimate before launch.
• turbopuffer: pro, very low storage and query costs; con, no free tier and a 10 KB minimum per write.
• Qdrant Cloud: pro, predictable resource-based billing and open source; con, you size and pay for the cluster even when idle.
• Weaviate Cloud: pro, Query Agent and built-in embeddings; con, the $45 Flex minimum and dimension-based billing that needs the calculator.
• Chroma Cloud: pro, simple per-GiB pricing and a familiar API; con, the jump to $250 a month for Team.

Verdict

There is no single cheapest vector database for agents; the right pick depends on how your agent reads and writes. For most new agent products with per-user memory, start on turbopuffer or Pinecone and batch your writes. Choose Qdrant when query volume is high and steady, Weaviate when you want the Query Agent, and Chroma when you are moving a local prototype to the cloud. Whatever you choose, estimate costs with your real write sizes, because agent memory writes are where list prices most often mislead.

Sources

• Pinecone pricing
• turbopuffer pricing
• Qdrant Cloud pricing
• Weaviate pricing
• Chroma Cloud pricing