PACINFRAX · RESOURCES

Choose the operating responsibility, not just the endpoint.

Compare four deployment paths by workload, cost unit and acceptance gates. This decision guide cannot invoke models, create endpoints, reserve capacity or purchase a commitment.

Planning only. None of these paths is an available production service. The local gateway is a deterministic mock; public draft model records are not its executable catalog.

Shared model API

Explore a model contract before committing dedicated capacity.

Who operates what

PacinfraX would operate serving; the customer controls project access and workload inputs.

Cost boundary

Per-model input/output units. Current draft text estimates exclude other charges.

Before activation

Licensed model revision, tested serving profile, region/data policy and approved rating.

Inspect draft model records →

Reserved throughput

Plan predictable demand when a measured throughput commitment is justified.

Who operates what

A future accepted contract must define eligible requests, burst policy and measurement windows.

Cost boundary

No PacinfraX throughput unit or reservation price has been approved. Token prices do not establish one.

Before activation

Measured capacity, explicit commitment/overage rules and auditable service objectives.

Prepare a local workload brief →

Dedicated model endpoint

Plan a controlled serving configuration for a specific model and workload.

Who operates what

Separate deployment configuration from observed readiness; isolation and scaling need independent validation.

Cost boundary

GPU/configuration time and ancillary services require an approved offer, not a public reference rate.

Before activation

Accepted inventory, runtime compatibility, provisioning, drain/rollback and failure recovery.

Review cloud responsibilities →

Customer-owned GPU deployment

Plan model serving on hardware you own, with optional hosting and managed operations.

Who operates what

Hardware title, supplier custody, customer workload rights and operator duties remain separate.

Cost boundary

Hosting, power, network and management are separate from inference; commercial terms are pending.

Before activation

Facility/custody evidence, power/cooling/network acceptance, rights and a reviewed service agreement.

Open local qualification rehearsal →

A decision passport follows every path.

Keep model revision, region/data policy, evidence freshness, price version, operating responsibility and unmet gates together. A completed plan does not make inventory available or authorize a billable request.

Hardware verification is scheduled for the final owner-provisioned integration phase. Identity, jobs, financial integrity, support and operator workflows are being developed independently of that purchase.

Read implemented local API boundaries →