AI & ML
Free Vector Databases and Embedding APIs: 11 Free Tiers Compared (2026)
Free vector database and embedding API tiers for developers: Pinecone, Qdrant, Weaviate, Zilliz, Upstash, Supabase, MongoDB, Voyage, Jina and Gemini compared.
For a free vector database that keeps running, Qdrant Cloud is the most predictable: a 1 GB RAM cluster, no card, and it holds about a million 768-dimension vectors. Weaviate Cloud and Zilliz Cloud also have free plans with no card, and if you already run Postgres, Supabase gives you pgvector inside a 500 MB project. For embeddings, Voyage AI gives 200 million free tokens on each of its current models, Jina has a free API key with a high rate limit, and Gemini's embedding model is free of charge on Google's free tier. Every number below was checked against the provider's own pricing or docs page on 13 November 2026.
Free tier or free trial
Two kinds of offer sit side by side here, and they suit different moments.
A free plan that stays free is what you want for a prototype, a side project or a demo that has to keep working: Qdrant, Weaviate, Zilliz, Upstash, Pinecone and Supabase work this way. The catch is inactivity. Qdrant suspends an unused free cluster after one week and deletes it after four weeks if it is not reactivated; Supabase pauses free projects after one week of inactivity; MongoDB pauses idle free clusters after 30 days with no connections. A free vector store is fine for a project you touch weekly and risky for one you forget about.
A one-time token grant is a budget, not a plan. Voyage AI's 200 million tokens is the large example. Spend it on evaluating models with your own data, not on a production workload you cannot pay for later.
Vector databases
Qdrant Cloud
The Qdrant Cloud free tier is a single-node cluster with 0.5 vCPU, 1 GB of RAM and 4 GB of disk. Qdrant's docs say that is enough for roughly one million 768-dimension vectors, and that you do not need a credit card to join. It also includes free cloud inference with selected models.
Good for: a real RAG prototype with a meaningful corpus, where you want a dedicated vector engine. Not for: anything you will leave untouched, because of the one-week suspension and four-week deletion described above.
Weaviate Cloud
Weaviate lists a free plan at $0 a month with no credit card required. It gives you one cluster per user, 100,000 objects, 1 GB of memory and 10 GB of disk, with one collection and up to three tenants. It includes 2,000 embedding requests a day and 1,000 Query Agent requests a month. Availability is best effort, with no backups or replication, on AWS in two regions.
Good for: learning Weaviate's hybrid search and its built-in embeddings. Not for: multi-collection schemas, since the plan allows one collection.
Zilliz Cloud (Milvus)
Zilliz Cloud is the managed Milvus service. The Free plan includes 5 GB of storage, 2.5 million vCUs per month and up to 5 collections, with community support. Zilliz's docs say no payment information is needed to create an account. Separately, signing up with a work email adds $100 in trial credits that expire after 30 days; adding a payment method earns another $100 and extends expiry to one year.
Good for: more storage than the other free clusters, and five collections. Not for: teams that need support beyond Slack and community help.
Pinecone
The Pinecone Starter plan gives 2 GB of storage, 2 million write units and 1 million read units a month, up to 5 indexes and 100 namespaces per index, in AWS us-east-1 only. It also includes 5 million embedding tokens a month with Pinecone's hosted inference. The pricing page does not say whether a card is required, and we did not confirm an inactivity policy.
Good for: the most polished serverless experience and the usage-based units model. Not for: any region other than AWS us-east-1.
Upstash Vector
Upstash Vector is serverless and priced per request. The Free plan allows 10,000 queries and updates a day, vectors up to 1,536 dimensions, 100 namespaces and 1 GB of total data, with no uptime SLA. The pricing page says a card is entered when you upgrade to pay-as-you-go but does not state whether the free plan needs one.
Good for: serverless and edge apps with light, bursty traffic. Not for: 3,072-dimension embeddings, which exceed the 1,536 cap.
Supabase (pgvector)
Supabase enables pgvector as a Postgres extension, so vectors live next to your normal tables, auth and row-level security. The Free plan has 500 MB of database per project, up to 2 active projects, and pauses free projects after one week of inactivity. The pricing page did not confirm card requirements or pgvector's availability on Free, but the docs describe the extension generally, not as a paid feature.
Good for: apps that need relational data and embeddings in one query. Not for: large embedding collections, where 500 MB goes quickly. If you query Postgres through an ORM such as Drizzle ORM, you keep one data layer.
MongoDB Atlas Vector Search
MongoDB's free Atlas cluster has 512 MB on the pricing page and 0.5 GB per the free-cluster limits, with a cap of 500 connections. The Vector Search limitations page caps Free clusters at 3 indexes in total, counting both search and vector indexes. We could not find an explicit statement that Vector Search is enabled on free clusters, so confirm it in your own cluster before you plan around it.
Good for: teams already storing documents in MongoDB. Not for: a vector-only workload.
Turbopuffer
Turbopuffer has no free tier on its pricing page. Its Launch plan has a $16 a month minimum usage and Scale has $256. We list it because it comes up in every comparison, not because you can use it for free.
Embedding APIs
Voyage AI
Voyage AI gives the first 200 million tokens free on every account for voyage-4-large, voyage-4 and voyage-4-lite. Older models such as voyage-3.5 get no free tokens. Our AI API offers record agrees on the 200 million figure. The pricing page did not state a card rule.
Jina AI
Jina lists embeddings limits of 100 requests and 100,000 tokens a minute without an API key, and 500 requests and 2,000,000 tokens a minute with a free API key. It is rate-limited access, not a token grant, so it suits steady low-volume use.
Google Gemini
On Google's pricing page, gemini-embedding-001 input is listed as free of charge on the free tier, and $0.15 per 1 million input tokens on the paid tier. Free-tier prompts may be used to improve Google products outside the EEA, UK and Switzerland, so keep private data off it.
Cohere
Cohere says calls from a Trial API key are free, but trial keys are rate limited and not permitted for production or commercial use. The page did not give a monthly call cap, so we do not quote one.
OpenAI
OpenAI does not offer a free tier for embeddings. text-embedding-3-small costs $0.02 per 1 million input tokens and text-embedding-3-large $0.13, which is the price to compare the free options against.
All of them side by side
| Provider | Type | What's free | Card |
|---|---|---|---|
| Qdrant Cloud | Vector DB | 1 GB RAM, 4 GB disk, ~1M 768-dim vectors | No |
| Weaviate Cloud | Vector DB | 100,000 objects, 1 collection | No |
| Zilliz Cloud | Vector DB | 5 GB, 5 collections | No (account creation) |
| Pinecone | Vector DB | 2 GB, 1M read units / month | Not stated |
| Upstash Vector | Vector DB | 10K ops / day, 1 GB | Not stated |
| Supabase | Postgres + pgvector | 500 MB, 2 projects | Not stated |
| MongoDB Atlas | Document DB | 0.5 GB, 3 indexes | Not stated |
| Voyage AI | Embeddings | 200M tokens per model | Not stated |
| Jina AI | Embeddings | 500 RPM with free key | Not stated |
| Gemini | Embeddings | Free of charge on free tier | Not stated |
| Cohere | Embeddings | Trial key, non-production | Not stated |
How to choose
- Biggest free corpus in a dedicated engine: Zilliz (5 GB) or Qdrant (about a million 768-dimension vectors).
- Vectors and relational data in one place: Supabase pgvector, or MongoDB if you already use it.
- Serverless, bursty traffic: Upstash Vector, remembering the 1,536-dimension cap.
- Free embeddings with no spend: Gemini for a rolling free tier, Voyage for a large one-time budget, Jina for rate-limited access.
- Keep the embedding model and the database dimensions in step: a 1,536 cap or a 768-dimension estimate only holds if your model produces that size.
For chat models on free tiers, see free AI API tiers that last, and for tracing what your RAG app does with them, LLM observability tools with free tiers. Open WebUI is a self-hosted chat interface with RAG if you want a front end to test retrieval. More offers are in the free tiers hub and the AI & ML directory.
Sources
Provider pages checked on 13 November 2026:
- Pinecone: pricing
- Qdrant: pricing, create a cluster
- Weaviate: pricing
- Zilliz: pricing, register
- Upstash: Vector pricing
- Supabase: pricing, vector columns
- MongoDB: pricing, free cluster limits, Vector Search limitations
- Turbopuffer: pricing
- Voyage AI: pricing
- Jina: pricing
- Google Gemini: pricing
- Cohere: pricing
- OpenAI: pricing