Peter Smith
While AI giants in the U.S. and China dominate headlines, a relatively unknown Australian company has just achieved something remarkable. Isaacus has built an AI model that outperforms both OpenAI and Google in the specialized world of legal research—proving that breakthrough innovation doesn't always come from the usual suspects.
A New Standard for Legal AI
A new Australian AI model is reshaping what's possible in legal intelligence. In a recent tweet, Wes Roth pointed out that Isaacus' newly launched Kanon 2 Embedder has claimed the top spot on the Massive Legal Embedding Benchmark (MLEB)—a respected standard for evaluating legal AI systems. According to the results, Kanon 2 beat OpenAI's Text Embedding 3 Large by 9% and Google's Gemini Embedding by 6%, all while running 30 to 340% faster. That combination of accuracy and speed is practically unheard of.
The MLEB chart, which measures model performance against inference time, places Kanon 2 squarely in the "new best" zone—meaning it delivers better retrieval accuracy at much lower latency. That's a big deal for real-time legal research tools and retrieval-augmented generation (RAG) systems. Meanwhile, well-known models like Voyage 3 Large, Gemini Embedding, and Text Embedding 3 Large sit in the "inferior models" region, balancing higher costs with less precise results.
Kanon 2 was trained on millions of legal documents spanning 38 countries, covering multiple jurisdictions, languages, and document types. That diversity gives it a serious edge in pulling up precedent-based or statutory information—something general-purpose language models often fumble due to legal nuances.
Privacy-First for Sensitive Work
Beyond raw performance, Kanon 2 offers a privacy-first deployment option, meaning it can run entirely on-premise to meet strict data-sovereignty requirements. That's huge for law firms, courts, and government agencies handling confidential information.
Experts see Kanon 2 as part of a bigger shift in AI: instead of building ever-larger general models, smaller domain-specific systems are proving more effective in fields like medicine, finance, and law.
Australia Joins the AI Big Leagues
Isaacus' success shows Australia is becoming a real player in the global AI race. By excelling in a niche but critical area like legal tech, the country proves that world-class innovation isn't limited to Silicon Valley or Shenzhen anymore. Kanon 2's efficient design and strong performance-to-cost ratio could make it the go-to model for next-gen legal copilots and RAG applications worldwide.
As legal systems lean more heavily on AI for document search, precedent analysis, and judgment summarization, Kanon 2 demonstrates that speed and precision can go hand in hand—and that the next big AI breakthrough might come from innovators working far outside Big Tech's orbit.
Peter Smith
