Leading Providers for Multimodal Embedding Storage & Search
The short answer
Pick the storage-and-search layer by where your data and your cloud contract already are; pick the embedding model by media type. As of November 8, 2026, the leading providers are:
- Amazon S3 Vectors — cheapest storage for very large multimodal corpora on AWS.
- Pinecone — best fully managed, cloud-neutral vector database.
- Weaviate — best for hybrid and multi-vector search, with an open-source core you can self-host.
- Google Agent Retrieval (formerly Vertex AI Vector Search 2.0) — best when you want embeddings generated for you on Google Cloud.
- Azure AI Search (Foundry IQ) — best for Microsoft estates and agents built in Foundry.
- Cloudflare AI Search — simplest managed pipeline from files to answers, with OCR included.
If you have not chosen an embedding model yet, start with our multimodal embedding model comparison; this page covers where those vectors live and how they are searched.
The comparison
| Provider | What you get | Price (list, Nov 8, 2026) | Built-in embedding | Best for |
|---|---|---|---|---|
| Amazon S3 Vectors | Vector buckets and indexes in S3, metadata filtering | Storage $0.06/GB-month · upload $0.20/GB · $2.50 per 1M queries + data processed | No (pair with Bedrock models such as Nova Multimodal Embeddings) | Billions of vectors, cost first |
| Pinecone | Serverless indexes; dense, sparse and full-text; Dedicated Read Nodes | Builder $20/month flat · Standard $50/month minimum: storage $0.33/GB-month, writes $4–$4.50/1M WU, reads $16–$18/1M RU · Enterprise $500 minimum | Yes (Pinecone Inference) | Managed, low-latency search on any cloud |
| Weaviate Cloud | Hybrid search, multi-tenancy, vector compression, Query Agent | Free (100K objects) · Flex from $45/month: vector dimensions from $0.00465/1M, storage from $0.12/GiB | Yes (Weaviate Embeddings, incl. ColModernVBERT) | Hybrid and multi-vector retrieval |
| Google Agent Retrieval | Collections of JSON data objects, auto-tuned ANN, auto-embeddings or bring your own | Usage-based or resource-based (see Google pricing estimate) | Yes (built-in models) | Google Cloud teams, Gemini Embedding 2 |
| Azure AI Search (Foundry IQ) | Vector + keyword + semantic ranker, agentic retrieval, image extraction | Dedicated tiers hourly; Serverless (preview) billed in compute-unit hours since Sep 13, 2026 | Integrated vectorization skills | Microsoft 365 / Foundry agents |
| Cloudflare AI Search | Managed ingestion, chunking, OCR, hybrid search over files and websites | Free until Dec 1, 2026; then 5M ingestion tokens, 10 GB-month and 1,000 + 1,000 queries included, $0.75 per 1M ingestion tokens (+$0.50 images), $2/GB-month, $0.75 per 1,000 semantic queries | Yes | Fastest path from documents to search |
Prices from each vendor’s pricing page or documentation on November 8, 2026. Embedding-model prices are on our current API prices table.
1. Amazon S3 Vectors — cheapest at scale
S3 Vectors stores embeddings in S3 itself, so storage costs object-storage prices rather than RAM prices. AWS lists $0.06 per GB-month for storage, $0.20 per GB uploaded, and queries at $2.50 per million plus a data-processed charge that falls from $0.004 per TB (first 100,000 vectors) to $0.0004 per TB above 10 million. In AWS’s worked example, 10 million 1,024-dimension vectors with 2 KB of metadata each fill 59 GB and cost $3.54 a month to store. Pair it with Amazon Nova Multimodal Embeddings in Bedrock, which AWS describes as one model for text, documents, images, video and audio.
Pick it if you have huge archives (video libraries, image catalogues, scanned records) and can accept slower queries than an in-memory database.
2. Pinecone — best managed vector database
Pinecone runs serverless indexes on object storage with dense, sparse and full-text indexes, and adds Dedicated Read Nodes for predictable query latency. The Standard plan has a $50 monthly minimum and bills storage at $0.33 per GB-month, write units at $4–$4.50 per million and read units at $16–$18 per million depending on cloud and region. A new Builder plan costs $20 a month flat with higher limits than Starter. Enterprise adds a 99.95% SLA, private endpoints, customer-managed keys and bring-your-own-cloud.
Pick it if you want a managed service on AWS, GCP or Azure with low query latency and no infrastructure to tune.
3. Weaviate — best for hybrid and multi-vector search
Weaviate combines vector and keyword search in one query, supports multi-tenancy and compression, and its cloud service includes embedding models such as ColModernVBERT ($0.065 per million tokens) for late-interaction visual document retrieval. Flex starts at $45 a month with a 99.5% SLA; Premium contracts reach 99.95% on AWS, GCP and Azure. The database is open source, so you can self-host the same engine.
Pick it if keyword matching matters alongside semantics, or you want multi-vector (per-patch) retrieval for documents and images.
4. Google Agent Retrieval — embeddings handled for you
Google renamed Vertex AI Vector Search 2.0 to Agent Retrieval. You store JSON data objects in collections, and the service can populate embedding fields automatically using built-in models or accept your own vectors, with filtering and auto-tuned indexes. Pricing is either usage-based for smaller workloads or resource-based for tuned performance.
Pick it if you are on Google Cloud and plan to use Gemini Embedding 2, which handles text, images, audio, video and PDF in one space.
5. Azure AI Search (Foundry IQ) — for Microsoft estates
Microsoft now sells Azure AI Search as the retrieval engine behind Foundry IQ knowledge bases. It combines vector, keyword and semantic ranking, supports multimodal embeddings, and offers add-ons for agentic retrieval and image extraction. Dedicated tiers bill hourly; a Serverless (preview) tier that scales to zero started billing on September 13, 2026.
Pick it if your agents live in Microsoft Foundry or Copilot Studio and your documents live in SharePoint and Azure storage.
6. Cloudflare AI Search — simplest pipeline
Cloudflare AI Search ingests files or crawls websites, chunks, embeds and indexes them, and serves semantic, full-text or hybrid queries. OCR is available on every account. Billing begins December 1, 2026: each account then gets 5 million ingestion tokens, 10 GB-month of storage and 1,000 semantic plus 1,000 full-text queries a month, and pays $0.75 per million ingestion tokens (an extra $0.50 for image processing), $2 per GB-month and $0.75 per 1,000 semantic queries. Limits to check: 1 million files per instance (500,000 with hybrid search) and 10 KiB of metadata per vector.
Pick it if you want search over documents running within days, without building ingestion yourself.
How to choose
- Corpus above ~100 million vectors and cost-sensitive: S3 Vectors.
- Need sub-100 ms queries on any cloud: Pinecone or Weaviate.
- Mixed keyword + semantic queries: Weaviate, Azure AI Search or Cloudflare AI Search (all hybrid).
- Committed to one hyperscaler: stay native (S3 Vectors, Agent Retrieval, Azure AI Search) so data never leaves the account.
- Small team, documents only: Cloudflare AI Search.
For general-purpose vector databases without the multimodal angle, see best vector databases.
Last verified: November 8, 2026.