Pricing
Simple pricing, no surprise compute bills
You bring the hardware. ZML charges for the orchestration layer, not per-token. Pick a tier that matches how many accelerators you're running.
Monthly
Annual
Save 20%
Maker
$0
Free, no credit card
- Up to 2 accelerator nodes
- 1 concurrent model deployment
- Unified REST and gRPC API
- Community support
- Basic Prometheus metrics
Most popular
Team
$
249
/mo
$
199
/mo, billed annually
Billed monthly
$2,388 per year, save $600
- Up to 20 accelerator nodes
- Unlimited concurrent deployments
- Latency and cost-aware routing
- OpenTelemetry tracing
- Team roles and access controls
- Email support
Production
$
849
/mo
$
699
/mo, billed annually
Billed monthly
$8,388 per year, save $1,800
- Unlimited accelerator nodes
- Multi-region deployments
- Priority routing and SLA options
- Custom routing policies
- SSO and audit logs
- Dedicated support channel
FAQ
Common pricing questions
A node is a physical or virtual host that ZML registers as an inference target. A single server with 8 x H100s counts as one node. A bare-metal AMD MI300X box counts as one node. Cloud VMs also count as single nodes. The node limit refers to how many registered inference targets ZML routes across simultaneously.
Yes. Upgrades take effect immediately. Downgrades take effect at the end of your current billing period. There are no penalties or setup fees for switching. Your configurations and deployment history are preserved across all tier changes.
No. ZML is priced on the orchestration layer, not on throughput. You pay a flat fee for the number of nodes you operate. There are no per-request, per-token, or per-GPU-hour charges from ZML. Your hardware costs are separate and handled by your infrastructure provider.
The Maker tier is permanently free with up to 2 nodes, which is enough to evaluate ZML on real hardware. For teams that need a longer evaluation period on a larger node count, reach out via the contact page and we'll set up a time-limited trial.
No. ZML runs entirely inside your own infrastructure. Model weights, inference requests, and outputs never leave your network. The only outbound connection is a telemetry ping for license validation and usage metering, which transmits node count and tier information only.
Start on the Maker tier today
Free, no card required. Connect up to two nodes and run your first multi-accelerator routing policy in minutes.