OPEN AND CLOSED · LLAMA · MUSE

Meta's Llama family brought capable open-weight language models to a wide audience, with sizes ranging from small models that run on a laptop to large multimodal models for servers. Llama weights are published under Meta's own community licence, which permits commercial use subject to conditions, rather than a standard open source licence.

In 2026 Meta Superintelligence Labs introduced the Muse family, which now powers the Meta AI assistant across Meta's apps and is offered to developers through Meta's own model API. Licensing differs between Muse releases, so check each one before building on it, and expect Meta's model line-up to keep changing.

For Australian businesses, Llama's main appeal is control. Because the weights can be downloaded, Llama can run entirely on servers in Australia with no third-party model provider in the data path, and it can be fine-tuned on proprietary data without that data leaving your environment.

PROCESSING IN AUSTRALIA

Onshore if you self-host

Self-host Llama weights on Australian servers

CHECKED 1 OCTOBER 2026 · CONFIRM BEFORE SENDING SENSITIVE DATA

001 / WAYS TO ACCESS

  1. 01Self-hostedDownload Llama weights and run them on Australian servers or private cloud GPUs.
  2. 02Microsoft FoundryLlama models as serverless Global Standard deployments, or on your own managed compute.
  3. 03Amazon BedrockSeveral Llama models in US and other regions, but not Sydney or Melbourne.
  4. 04Google Cloud Gemini Enterprise Agent PlatformLlama as managed open models, not listed for processing in Australia.
  5. 05Meta AI and the Meta Model APIMeta's own assistant and developer API, now centred on the Muse family.

KEEPING DATA IN AUSTRALIA

The reliable way to keep Llama processing in Australia is to self-host it: run the weights on GPU servers in an Australian data centre, or on virtual machines you control in AWS Sydney or Melbourne, Azure Australia East or Google Cloud Sydney. Managed serverless Llama is harder to keep onshore. Amazon Bedrock's model cards do not list Llama in Sydney or Melbourne, Microsoft Foundry offers serverless Llama only as Global Standard deployments, and Google Cloud does not list its managed Llama models for Australian processing. Confirm by checking each platform's regional availability for the exact Llama model, and if you self-host, document where weights, logs and backups are stored.

002 / STRENGTHS AND LIMITS

GOOD AT

  • +Weights can be downloaded and run on infrastructure you control
  • +Large ecosystem of fine-tunes, quantised versions and tooling
  • +Range of sizes, from on-device models to large multimodal models
  • +Supported by self-hosting tools such as Ollama, LM Studio and vLLM

WATCH OUT FOR

  • –The Llama community licence has conditions and is not an OSI-approved open source licence
  • –Amazon Bedrock does not offer Llama in its Australian regions, and Azure and Google Cloud host it on global endpoints
  • –Self-hosting larger Llama models requires significant GPU capacity and operational skill
  • –Meta's newest flagship assistant models are moving to the Muse family, so Llama's future roadmap is less certain

BEST FOR

  • Organisations that want full control by self-hosting in Australia
  • Fine-tuning a model on proprietary data
  • Offline, air-gapped or edge deployments
  • Prototyping locally before choosing a production model

003 / TOOLS BUILT ON IT

OTHER PROVIDERS

004 / QUESTIONS

Anything else is quicker to answer on a call. Thirty minutes, free, with the people who would build it.

Generally yes, under Meta's community licence, which allows commercial use but includes conditions such as an acceptable use policy and extra terms for very large services. Read the licence for the specific Llama release, because terms have varied between generations.

We choose the deployment that keeps your data where it needs to be, build the app around it and test it before launch.

Book a free call→