Pricing and Plan Structure in 2025

Vultr organizes instances into five tiers: Cloud Compute (shared vCPU), High Performance (AMD EPYC), High Frequency (NVMe, Intel), Optimized Cloud Compute (dedicated vCPU), and Bare Metal. The shared Cloud Compute starts at $2.50/month for 512 MB RAM and 10 GB SSD - useful for DNS or lightweight cron boxes. For anything running real queries or handling HTTP traffic, the High Performance AMD line is where the value is.

The 1 vCPU / 1 GB RAM High Performance instance costs $6/month. For comparison, DigitalOcean's equivalent Premium AMD Droplet is $7/month. Linode (now Akamai Cloud) sits at $6/month for the same spec but on older hardware in most regions. Vultr's High Frequency line at $8/month for 1 vCPU / 1 GB is harder to justify since the AMD EPYC instances outperform it in our fio tests.

One pricing trap: egress bandwidth. Each instance gets a monthly transfer allowance (1 TB on the $6 plan). Overage is $0.01/GB outbound, which is standard, but inbound is free. If you are running a mirror or a data pipeline pulling from S3, that helps. Vultr's bare metal starts at $120/month for an AMD EPYC 7443P with 256 GB RAM - genuinely competitive for teams that need single-tenant compute without colocating hardware.

# Compare monthly cost for 4 vCPU / 8 GB across providers (as of Q4 2025)
# Vultr High Performance AMD:  $24/month
# DigitalOcean Premium AMD:    $28/month
# Linode Dedicated CPU:        $30/month
# Hetzner CPX31 (EU only):     $13/month  <- still wins on raw EUR pricing

Benchmark Results: Storage and CPU

We used fio 3.35 for block storage, sysbench 1.0.20 for CPU, and pgbench against PostgreSQL 16.2 for a real-world OLTP simulation. All tests ran on High Performance AMD instances (4 vCPU / 8 GB) in the New Jersey region.

For storage, the random 4K read IOPS came in at 68,000 IOPS with a 0.23ms average latency. Sequential read hit 3,400 MB/s. The equivalent DigitalOcean Premium Droplet returned 48,000 IOPS random 4K read in the same test suite. The gap is real and repeatable across five separate runs. Write IOPS were closer: 42,000 on Vultr vs 38,000 on DigitalOcean.

For CPU, sysbench's prime number test (single-threaded) returned 3,812 events/second on the Vultr EPYC instance, consistent with an EPYC 7003-series core. PostgreSQL pgbench at scale factor 100 with 10 clients pushed 4,200 TPS on Vultr. On DigitalOcean Premium AMD with identical PostgreSQL configuration, we measured 3,650 TPS. That 15% lead matters if you are sizing a database server.

One caveat: these are shared instances, so results vary slightly under noisy-neighbor conditions. We saw a standard deviation of about 4% across runs, which is acceptable.

# fio command used for random 4K read test
fio --name=rand-read-4k \
  --ioengine=libaio \
  --iodepth=64 \
  --rw=randread \
  --bs=4k \
  --direct=1 \
  --size=8G \
  --numjobs=4 \
  --runtime=60 \
  --group_reporting \
  --filename=/dev/vda

Network: Latency, Peering, and IPv6

Vultr operates 32 locations as of late 2025. We tested latency from a home connection in London to three regions. Amsterdam returned 8ms average, New Jersey 88ms, Tokyo 242ms - all within expected ranges for transatlantic and transpacific hops.

IPv6 is dual-stack by default on all new instances. No configuration needed. You get a /64 per instance. This is table stakes in 2025, but some smaller providers still charge extra or require a support ticket. Vultr handles it correctly at provisioning time.

Intra-region private networking uses RFC 1918 space and is enabled per-instance. Traffic over the private interface does not count against your bandwidth quota. For a cluster communicating heavily between nodes - say, a Ceph storage pool or a Redis sentinel setup - this matters. We ran iperf3 across two instances on the same private network:

The throughput came in at 9.8 Gbits/sec, which is the practical ceiling for a 10Gbps virtual interface. No configuration was needed beyond assigning both instances to the same VPC network in the Vultr console.

# Private network iperf3 test between two instances in the same VPC
# On server node:
iperf3 -s -B 10.1.0.2

# On client node:
iperf3 -c 10.1.0.2 -t 30 -P 4

# Result: 9.81 Gbits/sec (4 parallel streams, 30 seconds)
// advertisement

FreeBSD and Operating System Support

Vultr is one of the few major cloud providers where FreeBSD actually works without fighting the platform. As of Q4 2025, Vultr offers FreeBSD 13.3 and 14.1 as first-class images. The cloud-init equivalent is handled via Vultr's own agent, but you can also use FreeBSD's built-in cloud-init support if you deploy a custom image via ISO upload.

We deployed a FreeBSD 14.1 jail host running bhyve as the hypervisor. The virtio network and storage drivers load cleanly. The Vultr kernel modules are not needed on FreeBSD - standard virtio_blk and vtnet handle everything.

ZFS root works if you use a custom ISO. The stock Vultr FreeBSD image uses UFS. To get ZFS root, upload a FreeBSD 14.1 ISO and install manually - the installer supports ZFS root since FreeBSD 13. We tested this and the resulting instance boots correctly with ZFS on root, GELI encryption on the data pool, and jails running nginx 1.26 and PostgreSQL 16.

For teams running mixed Linux and BSD infrastructure, Vultr is the practical choice. Hetzner does not offer FreeBSD images. DigitalOcean dropped FreeBSD support years ago. If your stack involves FreeBSD jails, bhyve, or you simply want to run a non-Linux production server without managing physical hardware, [Vultr](https://vultr.com/?ref=PLACEHOLDER) is the realistic option in 2025.

# Install FreeBSD pkg and set up a jail on Vultr FreeBSD 14.1
pkg install -y ezjail
ezjail-admin install
ezjail-admin create www 10.0.0.2
ezjail-admin start www
jexec www pkg install -y nginx
jexec www service nginx start

API, CLI, and Infrastructure Automation

Vultr's v2 REST API is complete and well-documented. Every resource - instances, VPCs, DNS, object storage, load balancers, Kubernetes - is manageable via API. Authentication uses a Bearer token passed in the Authorization header. The API returns standard JSON and uses cursor-based pagination.

The official CLI is vultr-cli, written in Go, available as a static binary. Version 3.3 is the current release as of late 2025. Installation is a single binary drop:

For Terraform users, the Vultr provider is maintained by Vultr itself and supports all major resources. The provider version 2.21.0 covers Kubernetes, VPC peering, and managed databases. We use it in our own infrastructure to provision test environments.

Teams building heavier automation pipelines around Vultr - triggered deployments, cost-aware scaling, or multi-cloud orchestration - should look at [taskbotshub.ai](https://taskbotshub.ai), which supports Vultr's API natively as an integration target and can automate instance lifecycle events without writing custom webhook glue.

One API limitation worth knowing: snapshot operations are asynchronous and do not support webhooks natively. You have to poll the snapshot status endpoint. In practice, a snapshot of a 25 GB disk takes 3-8 minutes. Our automation scripts use a 30-second polling interval with a 15-minute timeout before raising an alert.

# Install vultr-cli v3.3 on Linux x86_64
curl -LO https://github.com/vultr/vultr-cli/releases/download/v3.3.0/vultr-cli_v3.3.0_linux_amd64.tar.gz
tar xzf vultr-cli_v3.3.0_linux_amd64.tar.gz
mv vultr-cli /usr/local/bin/
export VULTR_API_KEY="your-api-key-here"

# List all running instances
vultr-cli instance list

# Create a High Performance AMD instance in New Jersey
vultr-cli instance create \
  --region ewr \
  --plan vhp-1c-1gb-amd \
  --os 1743 \
  --label prod-web-01

Managed Kubernetes and Databases

Vultr Kubernetes Engine (VKE) reached general availability in 2022 and has matured since. As of late 2025, it runs Kubernetes 1.31. The control plane is free - you pay only for worker nodes. Node autoscaling works and we tested it by firing a sustained load against an nginx Ingress and watching the cluster scale from 3 to 7 nodes over 4 minutes.

The kubeconfig is downloadable via API or CLI in one command. cert-manager, the Vultr CSI driver for persistent volumes, and the Vultr CCM for LoadBalancer services all install cleanly via Helm. We had a working cluster with persistent storage and an external LoadBalancer within 12 minutes of first provisioning.

Managed Databases cover MySQL 8, PostgreSQL 16, and Redis 7. The PostgreSQL managed service creates a primary and standby in the same region with automatic failover. Backups run daily with a 7-day retention on the base plan. Connection pooling via PgBouncer is built in - you connect to port 5432 for direct access or port 5433 for the pooled connection.

One limitation: no cross-region read replicas for managed PostgreSQL as of our testing. For globally distributed reads, you need to manage replication yourself or use a different provider. Vultr's managed database pricing is also higher per-unit than self-managed on a High Performance instance. A managed PostgreSQL with 2 vCPU / 4 GB costs $60/month. A self-managed instance with equivalent specs is $18/month. You are paying $42/month for the managed overhead. Reasonable if you want automated failover and backups without the operational work, but sysadmins running their own backup and HA tooling should self-manage.

# Provision VKE cluster and get kubeconfig
vultr-cli kubernetes create \
  --label prod-k8s \
  --region ewr \
  --version 1.31.1+1 \
  --node-pools "quantity:3,plan:vhp-2c-4gb-amd,label:workers"

# Fetch kubeconfig once cluster is active
vultr-cli kubernetes config  > ~/.kube/vultr-prod.yaml
export KUBECONFIG=~/.kube/vultr-prod.yaml
kubectl get nodes
// advertisement

Object Storage and Block Storage

Vultr Object Storage is S3-compatible. Buckets are created per-cluster (not per-region - each storage cluster covers a geographic area). The endpoint format is https://.vultrobjects.com. It works with standard S3 clients: aws-cli, s3cmd, rclone, and the boto3 SDK.

Pricing is $0.02/GB/month for storage and $0.01/GB for outbound transfer. For reference, AWS S3 in us-east-1 is $0.023/GB/month storage and $0.09/GB egress. Vultr's egress rate is 9x cheaper than S3. For teams storing and serving large datasets or backups from Vultr compute, this difference is material at scale.

Block storage (NVMe) attaches as an additional volume to any instance. Maximum size is 10 TB. We tested a 500 GB block volume attached to a High Performance instance and got 2,800 MB/s sequential read in fio, which is slightly below the boot volume but within expected range for an attached volume. Block storage volumes survive instance deletion, making them suitable for persistent data that needs to outlive ephemeral compute.

One operational note: block volumes do not support live resize. You must detach, resize via API, reattach, then run resize2fs or growpart. Plan your initial volume sizes with growth in mind to avoid maintenance windows.

# Configure rclone for Vultr Object Storage
cat >> ~/.config/rclone/rclone.conf << 'EOF'
[vultr-s3]
type = s3
provider = Other
access_key_id = YOUR_ACCESS_KEY
secret_access_key = YOUR_SECRET_KEY
endpoint = https://sjc1.vultrobjects.com
acl = private
EOF

# Sync a local directory to a Vultr bucket
rclone sync /var/backups/ vultr-s3:my-backup-bucket/server-01/ --progress

Support, SLA, and Real Reliability

Vultr offers 99.99% uptime SLA for compute instances. In six weeks of monitoring with Prometheus and Blackbox Exporter pinging our test instances every 30 seconds, we recorded zero downtime across all three regions. This is one data point, not a longitudinal study, but it tracks with our longer-term experience running production workloads on Vultr since 2022.

Support is ticket-based with no phone option on standard plans. Average first response time during our tests was 47 minutes for low-priority tickets and 18 minutes for high-priority. The support team answered specific kernel and networking questions accurately without routing us to generic documentation. That matters for sysadmins who are not asking beginner questions.

Vultr does not offer a managed security service or DDoS mitigation beyond basic volumetric filtering at the network edge. If you need advanced DDoS protection, you will be fronting your infrastructure with Cloudflare or a dedicated scrubbing service regardless of cloud provider. Plan accordingly.

For project naming and DNS when spinning up new infrastructure on Vultr, if you need to register a clean domain name for a service or internal project, [nicename.me](https://nicename.me) is worth checking before defaulting to GoDaddy - the search interface is faster for evaluating available names across TLDs.

# Prometheus blackbox exporter config for instance availability monitoring
scrape_configs:
  - job_name: vultr_instances
    metrics_path: /probe
    params:
      module: [icmp]
    static_configs:
      - targets:
          - 45.76.100.10   # nj-prod-01
          - 95.179.200.20  # ams-prod-01
          - 108.61.150.30  # tyo-prod-01
    relabel_configs:
      - source_labels: [__address__]
        target_label: __param_target
      - target_label: __address__
        replacement: localhost:9115