Reserved for you. Operated for you.
Move a production workload onto capacity sized around its traffic, prompt shape, and performance target. We deploy the model, configure the serving stack, and operate it with you.
Reserved for your workload
Capacity is sized against your traffic profile and allocated to your deployment rather than an anonymous serverless queue.
Managed end to end
We deploy, configure, monitor, scale, and update the model so your team can stay focused on the product using it.
Limits you set
Per-key RPM and TPM ceilings are yours to allocate across teams, not a shared pool.
Autoscale with a floor
Set a minimum replica count for latency and a ceiling for cost. Burst overflow can spill to shared capacity or fail closed.
Sized with evidence
We measure your context, output length, concurrency, and target latency before recommending a deployment shape.
Named technical owner
A direct escalation path covers deployment changes, model updates, rollback planning, and production incidents.
Sizing
| Tier | Accelerators | Fits | Throughput | Price |
|---|---|---|---|---|
| S | 1× H200 | 8–30B dense, or a LoRA fleet | ~90 tok/s/stream | $3.20 / GPU-hr |
| M | 4× H200 | 70–120B dense, MoE up to 235B | ~140 tok/s/stream | $12.40 / GPU-hr |
| L | 8× H200 (NVLink) | 400B–1T MoE at BF16/FP8 | ~210 tok/s/stream | $24.60 / GPU-hr |
| XL | Multi-node, reserved | Custom weights, long-context serving | Sized with you | Annual commit |
Hardware, throughput, and rates are planning examples. Your benchmark determines the proposed deployment and commercial terms.
Workload proof
Test your traffic, not a headline benchmark.
Public catalog figures are planning estimates until a reproducible measurement is published. For production sizing, we run your prompt shape, output length, and concurrency target on the model and deployment you are considering.
Request a benchmark →p50 and p95 time to first token
Output tokens per second
Concurrency and tail-latency behavior
Estimated monthly serving cost
Recommended deployment size
Base-versus-tuned comparison, when supplied
{
"model": "llmcloud/llama-4-maverick",
"provider": {
"endpoint": "dedicated:acme-prod",
"region": "eu-west",
"fallback": "shared"
},
"messages": [{"role":"user","content":"hi"}]
}When is dedicated a better fit than serverless?+
When predictable capacity, customer-specific limits, tuned weights, or a defined performance target matters more than pure pay-as-you-go flexibility. The benchmark establishes the break-even point for your workload.
Can I run a model that isn't in the fleet?+
Any open-weight model we can legally serve, plus your own fine-tunes. Custom weights start at the M tier.
Does the gateway still fail over?+
Yes, if you allow it. Set provider.fallback to shared capacity or to third-party providers, or leave it unset to fail closed.
Which plan includes it?+
Managed dedicated inference is an Enterprise offer. Every tier can still call the hosted fleet per token; reserved capacity and managed operation are scoped separately.
How do I start?+
Request a benchmark and tell us the model, traffic shape, context and output length, and the performance target you need to meet.