Australian LLM inference · questions & answers
Questions about Australian LLM inference.
Direct answers about inference regions, API commitments, data boundaries and how to read the evidence in this directory.
Answers reflect the directory source review on 9 November 2026. Provider documentation and commercial terms can change; check the linked profile and source before making a deployment decision.
What does Australian LLM inference mean in this directory?
It means official, region-specific documentation supports Australian model inference for a named model and deployment. That claim does not automatically cover routing, moderation, logging, storage, support access or connected features; review those boundaries separately.
Which LLM APIs document model processing in Australia?
As of 9 November 2026, 7 named model configurations in this directory meet its “AU processing documented” evidence standard. Each is listed below with the model, deployment and official sources. This is a source review, not a guarantee of your account's live route or contract.
Does Australian data storage mean prompts are inferred in Australia?
No. Data storage location and model processing location are separate controls. A provider may store API data in Australia while processing prompts elsewhere. Check the precise model, deployment, feature and route rather than relying on a general regional-storage statement.
Does a Sydney API endpoint guarantee that LLM inference stays in Australia?
No. An endpoint name or IP address alone does not establish where a model processes a request. Global routing, cross-region inference and feature-specific handling can change the boundary. Look for current documentation for the exact model and deployment, and check whether fallback routes are enabled.
Can I use an Australian LLM API without reserving compute?
The directory prioritises shared, managed API options that do not require dedicated or reserved capacity. That does not mean every option is self-service or has no minimum spend, subscription, platform fee or account prerequisite. Check the separate access, billing and capacity fields on each profile.
What does ‘AU processing unverified’ mean?
It means the sources reviewed do not establish that the named service processes model requests in Australia. It is an evidence gap, not a finding that requests necessarily leave Australia. The directory keeps this separate from documented, conditional and offshore processing.
Does a no-training commitment mean an API does not retain prompts?
No. Training use and retention are different questions. Review logging, abuse monitoring, support access, backups, caches, stateful features and the terms for your account. A no-training statement by itself does not establish zero retention.
Does an Australia–New Zealand processing boundary mean Australia-only processing?
No. A combined Australia–New Zealand boundary allows processing in more than one country. The directory labels those Databricks configurations as conditional and does not include them in the Australia-only processing-documented filter.
What should I check before sending production prompts to an API?
Confirm the exact model and version, deployment type, region, routing and fallback behavior; then review storage, retention, abuse monitoring, support access, account eligibility, quotas, pricing and contract terms. Test the full request path using synthetic or approved data before production use.
What can a team do when no Australian route has the model or capacity it needs?
First define the model capability, throughput, latency and data constraints, then test whether a smaller model or a different workload design fits. If considering an upstream route outside Australia, assess its terms and processing boundary. Made Veil can redact supported identifiers before forwarding a request to a compatible API, but it does not add Australian compute or guarantee complete detection.
Who maintains the directory, and does a listing certify a provider?
Made maintains this source-linked research directory. Inclusion is not a paid placement or endorsement, and the review is not an independent audit, contractual guarantee or live API test. Each claim links to its source and is scoped to the evidence available on the review date.