Large Language Model Landscape in Australia: Technical Assessment

Key points

• There is no flagship race yet, and competition is fierce around the world, locally developed LLMs (such as GPT-4, Claude 3.5, Llama 3.1) have not yet appeared from Australia. Australian research and business currently rely mainly on international LLM, frequently used, but has measurable restrictions on Australian English and cultural background.
• Kangaroo LLM is the only major open source, locally developed LLM project. Supported by a consortium of Katonic AI, Rackcorp, NextDC, Hitachi Vantara and Hewlett Packard Enterprise, it aims to build a model specifically targeting Australian English, but is still in the early stages of data collection and governance, with no public model weights, benchmarks or production deployments as of August 20225.L.
• International Model (Claude 3.5 sonnet, GPT-4, Llama 2) is widely used in Australia for research, government and industry. Their deployment in an Australian environment is often affected by challenges in data sovereignty, privacy laws and model fine-tuning.
• Australian academic research has made important contributions to LLM assessment, equity and field adaptation, not infrastructure. Work at UNSW, Macquarie and Adelaide University focuses on bias detection, medical and legal applications, and fine-tuning of pre-trained models rather than building new large LLMs from Scratch.
• Governments and industries are investing in AI, but AI sovereignty remains aspiring. There are active policy developments, increased venture capital and strategic university-industrial partnerships, but there is no national computing infrastructure or business ecosystem for large-scale training of large general purpose LLMs.

Local model development: Kangaroo LLM

Kangaroo LLM It is Australia’s flagship effort to establish a sovereign, open source big-word model tailored to Australian English and culture. The project is managed by a nonprofit alliance to create a model for understanding Australian humor, language and legal/ethical norms. However, as of August 2025, Kangaroo LLM is Not fully trained, benchmarked or publicly available models. Its current state is best as follows:

• Partners: Katonic AI (Lead), Rackcorp, NextDC, Hitachi Vantara, Hewlett Packard Enterprise.
• mission: To create an open source LLM trained in Australian web content, data sovereignty and local cultural consistency are the main goals.
• progress: The project has identified 4.2 million Australian websites to collect potential data, with initially focusing on 754,000 sites. The crawl in late 2024 was postponed due to legal and privacy concerns, and no public data sets or models have been released.
• Technical Methods: The “Kangaroo Bot” crawler respects Robots.txt and allows exit from the site. The data is processed into the “Vegemighty dataset” and is refined through the “big barrier pipeline” used for LLM training. The architecture, size and training method of the model are still not disclosed.
• Governance: Work as a volunteer for a nonprofit organization (about 100 volunteers, more than 10 full-time equivalents). Business clients and possible government grants seek funding, but large amounts of public or private investment have not been announced.
• schedule: The project was originally scheduled to be released in October 2024, but as of August 2025, the project is still in the data collection and legal compliance phase, with no confirmed release date for the trained model.
• significance: Kangaroo LLM is a symbolic and practical step towards AI sovereignty, but it has not yet represented a technological alternative to global LLM. Success will depend on the ongoing funding, technology execution and adoption of Australian developers and businesses.

International model deployment

Claude 3.5 sonnet (Human), GPT-4 (Openai) and Llama 2 (Meta) is all available and actively used in Australian research and industry. Their adoption is driven by their outstanding capabilities, easy-to-access drivers accessed through cloud providers (AWS, Azure, Google Cloud) and the drivers integrated into enterprise workflows.

• Claude 3.5 sonnet AWS’s Sydney region has been available since February 2025, enabling Australian organizations to use state-of-the-art LLMs with data residency compliance. This model is used for applications from customer service to scientific research.
• GPT-4 and Llama 2 It is widely used in Australian universities, startups and companies for prototypes, content generation and task automation. Their use is often accompanied by fine-tuning of local datasets to improve relevance and accuracy.
• University of Sydney Case Study: A team used Claude to analyze whale acoustic data and achieved 89.4% accuracy in detecting Minke whales, which is a significant improvement over the traditional method (76.5%). The project demonstrates how to adapt to local scientific needs, but also emphasizes Australia’s reliance on external model providers.

Research Contributions

Australian academic institutions engage in LLM research activities, but their focus is Assessment, equity, domain adaptation and application– Not to build new large-scale basic models.

• UNSW’s Besstie Benchmark: A systematic framework for emotional and ironic assessment of Australian, British and Indian English. It shows that global LLM has always performed poorly in Australian English, especially ironic detection (Reddit’s F-Score 0.59, while emotional 0.81). This work is essential to understand the limitations of the current model in the local environment.
• Biomedical LLM at Macquarie University: The researchers had the fine-tuned medical problems with the BERT variant (Albert Biobert, Albert) and scored the highest score in international competitions. This shows Australia’s strength in adapting existing models into areas of expertise, but not in developing new architectures.
• CSIRO DATA61: Use LLM, privacy protection AI and model risk management to conduct influential research on agent-based systems. Their work is practical, with a policy-focused rather than focusing on basic model development.
• University of Adelaide and Business Partnership: Founded at the end of 2024, Commbank Basic AI Center aims to advance machine learning in financial services, including fraud detection and personalized banking. This is a major industry investment, but again, the focus is on applications and fine-tuning rather than building new large LLMs.

Policy, investment and ecosystem

Government Policy:
The Australian government has developed a risk-based AI policy framework and mandatory transparency, testing and accountability for high-risk applications. The 2024 privacy law reform introduces new requirements for AI transparency, affecting the selection and deployment of models.

invest:
Venture capital for AI startups in Australia reached US$1.3 billion in 2024, with AI accounting for nearly 30% of all venture capital transactions in early 2025. However, most of this investment is in application-level companies, not basic model development.

Industry adoption:
A 2024 survey found that 71% of Australian university employees used Generative AI tools, mainly Chatgpt and Claude. Business adoption is growing, but is often limited by data sovereignty requirements, privacy compliance, and the lack of locally tailored models.

Computing Infrastructure:
Australia does not have a large-scale sovereign computing infrastructure for LLM training. Most large model training and inference rely on international cloud providers, although AWS’s Sydney region now supports Claude 3.5 sonnets.

Summary

LLM Landscape in Australia by Strong application-driven research, enterprise adoption of growing and aggressive policy developmentbut No sovereignty, large-scale basic model. Kangaroo LLM is one of the few important efforts in the area, but it is still in its early stages and faces significant technical and resource barriers.

All in all, Australia is the sophisticated user and adapter of LLM, but it is not the builder of it yet. The most important element is clear: Kangaroo LLM is a meaningful step, but it has not been resolved. Global models dominate but have local restrictions; Australian research and policies are world-class in terms of evaluation and application, rather than basic innovation.

Source:

20. /kangaroo-bot/