APIbenchmarks
Zilliz Cloud (Milvus) logo

Zilliz Cloud (Milvus)

Zilliz · Ranked #3 of 7 in Vector Database APIs

85.4/ 100
AExcellent

Fully managed Milvus with a generous permanent free tier, 99.95% SLA, and massive open-source lineage.

Best for

Managed Milvus at billion-vector scale

Screenshot of Zilliz Cloud (Milvus)

Overview

Zilliz Cloud is the fully-managed, commercial vector database service built by the original creators of Milvus, the open-source vector database (40K+ GitHub stars). The relationship matters: Milvus is the engine, and Zilliz Cloud is the hosted, operationalized version that removes the considerable Kubernetes/distributed-systems burden of self-hosting Milvus. The platform targets teams building RAG pipelines, semantic search, recommendation systems, and AI agents at scale, offering deployment models from a free serverless tier up to dedicated clusters, Bring-Your-Own-Cloud (BYOC), and a Business Critical plan for regulated enterprises. It is available across AWS, GCP, and Azure, with SOC 2 Type II, ISO 27001, GDPR, and HIPAA-readiness positioning it for enterprise procurement.

Where it wins: raw performance and scalability. Milvus consistently ranks at or near the top in third-party and first-party VectorDBBench/VDBBench benchmarks, and Milvus 2.6 introduced RaBitQ 1-bit quantization delivering roughly 72% memory reduction with ~4x throughput gains while holding ~95% recall, plus 3-7x higher QPS than Elasticsearch on full-text workloads. Users routinely cite sub-200ms search across millions of vectors, billion-scale capacity, a rich index family (HNSW, IVF, DiskANN, GPU indexes), hybrid dense+sparse search, and broad SDK coverage. The managed service earns strong marks for ease of onboarding and documentation, and G2 named Zilliz "Highest Performer" and "Easiest to Use" in its Summer 2025 Grid for Vector Databases.

Where it loses: cost predictability and operational nuance at the edges. The most common complaint is price, for small or hobby projects, several reviewers note that self-hosting Milvus on a cheap VPS (e.g., Hetzner) or a mini-PC can be dramatically cheaper than Zilliz Cloud's managed bill, and the vCU/CU-hour-plus-storage model can be hard to forecast. Open-source Milvus itself carries a steep learning curve in distributed deployments (Kubernetes operators, dependent services, monitoring/backup per component), and schema migrations are described as complex and time-consuming. So the calculus is the classic managed-vs-self-hosted tradeoff: Zilliz Cloud is compelling for teams that value velocity and don't want to operate distributed infrastructure, less so for cost-sensitive small workloads.

How this score is derived

The APIbenchmarks Index is a weighted sum of four dimensions, each scored on an absolute 0–100 reference scale. See the methodology for every mapping.

DimensionScoreWeightContribution
Documentation & DXZilliz/Milvus docs are comprehensive with per-language SDK references, quickstarts, and tuning guides, and reviewers repeatedly cite the documentation as a reason onboarding was straightforward.
82
30%24.6
ReliabilityZilliz Cloud Dedicated publishes a 99.95% uptime SLA (with 99.9% for some on-demand/read-node configurations) backed by service credits, plus multi-replica and global-cluster region-level resilience options.
88
25%22.0
Ecosystem & SDKsBuilt on the most popular open-source vector DB (Milvus, 40K+ stars) with first-party integrations across LangChain, LlamaIndex, major embedding providers, AWS/GCP/Azure marketplaces, and an MCP server.
85
25%21.3
AccessibilityA genuinely free serverless tier (5GB storage + monthly vCU allowance), one-click signup, REST + gRPC APIs, and SDKs across Python/Java/Go/Node make it easy to start without infrastructure setup.
88
20%17.6
APIbenchmarks Index (ABI)85.4

Table 1. Derivation of the ABI for Zilliz Cloud (Milvus). Contribution = score × weight; the index is their sum.

At a glance

Vendor
Zilliz
Pricing model
Per compute-unit-hour + storage
Free tier
5GB storage + compute credits, permanent free tier
Official SDKs
9 languages

Pricing

Free (Serverless)$0/monthFree serverless tier including ~5GB storage and a monthly vCU allowance; suitable for prototyping. Collections unlimited.
ServerlessPay-as-you-go (from ~$0.35/M vCU)Pay-per-operation model charged on read/write resource consumption plus storage; best for variable/low-traffic workloads.
DedicatedFrom ~$99/month (compute + storage)Reserved compute units (CU ~= 1 vCPU + 4GB RAM) at pay-as-you-go rates; 99.95% uptime SLA, production-grade performance and isolation.
Business Critical (BC)CustomAdded Oct 2025 for teams with stringent security/compliance needs; enhanced isolation, support, and resilience.
BYOC (Bring Your Own Cloud)CustomZilliz Cloud control plane operating inside the customer's own cloud account/VPC for data-residency and compliance requirements.

Key features

  • Billion-scale approximate nearest-neighbor (ANN) vector search
  • Multiple index types: HNSW, IVF variants, DiskANN, GPU-accelerated indexes
  • RaBitQ 1-bit quantization (Milvus 2.6) for ~72% memory reduction
  • Hybrid search combining dense and sparse vectors
  • Built-in full-text / BM25 search
  • JSON path indexing with sub-millisecond filtered latency
  • Serverless, Dedicated, BYOC, and Business Critical deployment models
  • Multi-cloud (AWS, GCP, Azure) availability
  • Global Cluster option for region-level resilience and failover
  • Enterprise security: RBAC, audit logs, SOC 2 / ISO 27001 / HIPAA-ready, data resilience/backup

Official SDKs

Python (PyMilvus)JavaGoNode.js / JavaScriptC# (.NET, contributed by Microsoft)C++Rust (in development)RESTful APIgRPC API

Strengths & trade-offs

Strengths
  • +Top-tier benchmark performance: Milvus 2.6 delivers ~4x throughput and 72% memory reduction via RaBitQ quantization while holding ~95% recall, and 3-7x higher QPS than Elasticsearch on full-text workloads
  • +Scales to billions of vectors; reviewers report searching millions of vectors in ~200ms
  • +Removes the heavy operational burden of self-hosting distributed Milvus (Kubernetes, monitoring, backups, HA)
  • +Rich index and search capabilities: HNSW, IVF, DiskANN, GPU indexes, plus hybrid dense + sparse and full-text search
  • +Strong enterprise posture: 99.95% SLA, SOC 2 Type II, ISO 27001, GDPR, HIPAA-readiness, RBAC, BYOC, audit logs
  • +Genuinely free serverless tier and easy signup lower the barrier to start
Trade-offs
  • Cost is the most cited complaint, managed pricing can be far more expensive than self-hosting Milvus on a cheap VPS for small/hobby projects
  • vCU / CU-hour + tiered-storage billing model can be hard to forecast and reason about
  • Underlying Milvus has a steep learning curve, especially for distributed/self-hosted setups
  • Collection schema modifications require a migration process that is complex and time-consuming
  • Combining vector search with structured filtering can require application-level joins, adding latency/complexity
  • Vendor lock-in concerns and dependency on a cloud-based platform where instability interrupts access

What developers say

G2 4.7/5

Strongly positive overall, users praise performance, scalability, ease of use, and documentation, with cost for small projects being the dominant recurring criticism.

What I liked most about Zilliz was the ease-of-use. The implementation was straight forward and the documentation was really helpful.

Key figures

Query throughput (Milvus 2.6, RaBitQ)946 QPS vs 236 QPS baseline (~4x) at 94.9% recallMilvus blog (2.6 preview)
Memory reduction (Milvus 2.6 RaBitQ vs IVF_FLAT)72% reductionMilvus blog (2.6 preview)
Full-text search throughput vs Elasticsearch3-4x higher (up to 7x QPS on some workloads), equivalent recall, BEIR datasetMilvus blog (2.6 preview)
JSON path index P99 latencyReduced from 480ms to 10ms at 100M+ recordsMilvus blog (2.6 preview)
Uptime SLA (Zilliz Cloud Dedicated)99.95% (99.9% for some on-demand/read-node configs)Zilliz Cloud SLA
Dedicated compute price~$0.096 per CU-hour (~$69/mo per CU running 24/7)LeanOps pricing breakdown
Standard SSD storage price~$0.02 / GB / monthLeanOps pricing breakdown

Compare Zilliz Cloud (Milvus) head to head

Sources

  1. https://zilliz.com/pricing
  2. https://zilliz.com/sla
  3. https://zilliz.com/blog/zilliz-cloud-oct-2025-update
  4. https://leanopstech.com/blog/milvus-zilliz-cloud-pricing-2026/
  5. https://www.g2.com/products/zilliz/reviews
  6. https://www.g2.com/products/milvus/reviews
  7. https://milvus.io/blog/milvus-26-preview-72-memory-reduction-without-compromising-recall-and-4x-faster-than-elasticsearch.md
  8. https://github.com/zilliztech/VectorDBBench
  9. https://www.prnewswire.com/news-releases/zilliz-named-highest-performer-and-easiest-to-use-in-g2s-summer-2025-grid-report-for-vector-databases-302503076.html

Figures last verified 2026-06-27. Spotted an error? corrections@apibenchmarks.com