Pricing
Priced against productive GPU capacity, not seats.
Per GPU per year for a fleet, per box for a workstation, with the control plane and the economics layer included. Prices are set with you for now: tell us the fleet and we price it, and the payback model below shows what it is worth.
Contact us for pricing
Available now
Open source engine
The engine and every recipe, free. For developers, labs and anyone running open models on hardware they own.
- Metrale Engine, full source
- Every model recipe in the registry
- OpenAI, Anthropic and Responses APIs
- LAN fleet manager, early access
- Community support in Discord
Workstation and edge
For teams and organizations on local devices: a DGX Spark or Strix Halo class box in an office, a branch or the field. Signed update channel, managed from the console.
- Signed stable and LTS channels
- Console access for every licensed box
- Email support with a response SLA
- Volume pricing from 25 boxes
Most fleets start here
Enterprise
The full platform for GPU fleets. Support and forward deployed engineering are priced separately.
- Metrale Engine, signed stable and LTS channels
- Metrale Control, rollouts, routing, policy, repair
- Metrale Economics, chargeback and payback
- Named engineer and response SLA
- Hosted, your cloud, on premises or air gapped
Proof of value
One model, one hardware target, one workload. A side by side ladder in week one and a receipt in dollars per workload at the end.
- Scoped success criteria, yours or ours
- Side by side against your current engine
- Economics baseline of the target cluster
- Forward deployed engineer for the four weeks
- Fee credited against the first year on conversion
What the platform meters
Priced on a number you can watch.
Proposed. The platform is being built to meter what it serves and show it live, so a renewal is read off the same number the console shows.
GPU hours and GPU count
A license is a cluster or a number of GPUs, discovered by the platform, not declared on a form.
Tokens per GPU second
The throughput the fleet actually produced, per GPU, per second, next to the throughput it could have.
Cost per million tokens
GPU, storage, network and platform cost over the tokens delivered, for your fleet and for the baseline you ran before.
Nothing through the gateway without a license
Every served request is entitled and counted, so the bill and the telemetry are the same record.
Payback
Find your payback period.
Payback is what a thing costs over what it saves each month, and everything after is upside. That is the number to walk to the CFO with. Three scenarios, every input editable, evidence class on every field. The license is priced with you, so its field starts empty: put your quote in to see the payback.
The uplift frees GPUs. Freed GPUs are deferred purchases or rentals plus the power they burned. The license, priced with you, is what the uplift costs.
Uplift defaults to 1.20x, below the measured ratio on the GB10 ladder at C=128, because a datacenter part is not a DGX Spark until we publish the receipt.
- GPUs freed by the uplift
- 42.7
- Deferred purchase or rental
- $1,706,667
- Power no longer burned
- $40,815
- Gross savings per year
- $1,747,481
- Metrale license per year
- On quote
A model, not a quote. Savings depend on your workload, your utilization and the uplift measured on your hardware during the pilot. Evidence classes: MEASURED from ladder.generated.json, the published concurrency ladder. USER yours to edit, the model recomputes as you type.
Questions
The questions we actually get asked.
Short answers. Each one is backed by something on this site or in the repository.
How is it priced?
Per GPU per year for the Enterprise tier, with volume tiers as the fleet grows, and a per box price for workstation and edge deployments. Support and forward deployed engineering are priced separately. The engine itself is free and open source under MIT OR Apache-2.0. Prices are set with each customer for now, so the pricing page says contact us, beside a payback model with editable inputs.
How do you measure savings?
Against your own baseline. Economics records what each cluster cost per workload before Metrale takes traffic, then reports the delta as traffic moves. The pilot ends with a receipt in dollars per million tokens and dollars per successful workload, not a slide.
What is the payback period?
It depends on your fleet, your utilization and the uplift we measure on your workload. The model on the pricing page computes it from inputs you control. Every month after payback is upside, which is why we talk about payback rather than a percentage.
What about SOC 2 and compliance?
The architecture is built for regulated buyers, and SOC 2 readiness documentation, model risk documentation and pinned recipe governance packs are part of the first SLA engagements. Ask for the current state of the audit program when you book. We will tell you exactly where it is.
What support comes with it?
Community support in Discord for the open source engine. Workstation and edge includes email support with a response SLA. Enterprise includes a named engineer, a response SLA and a shared channel. Forward deployed engineering for the pilot and the cutover is scoped per engagement and credited against the first year on conversion.
Ask the rest in a working session, or read the deployment guide ↗.
Next step
Get a price, or get the receipt.
Email [email protected] and we price your fleet, or book a working session and we run the ladder on your workload.