Canberra scaleup Trellis Data’s sovereign AI model is now up and running for business and government

Résumé détaillé

Trellis Data describes its sovereign AI Agentaus 1 as “near-frontier” and claims it sits 43rd globally among foundation models.

Agentaus 1 has 220 billion parameters – the internal settings an AI model adjusts during training – about 50% more than its predecessor.

The Canberra deep tech scaleup says it gives more accurate answers, can work through problems that take several steps and produces longer responses. The model can also run on smaller clusters of computers, which the company says keeps costs down.

Trellis describes Agentaus 1 as “near-frontier” and claims it sits 43rd globally among foundation models.

Agentaus launched as a commercial service in May. Public servants can use it to turn policy submissions into a ministerial brief, with references back to the originals. Trellis says its Australian data centres handle both the documents and the processing, and it does not use customer material to train its models.

Michael and Rachel Gately founded Trellis Data in Canberra in 2016. Gately was working at a federal agency where a promising algorithm stalled because senior management saw an internal project as too risky.

By continuing, you agree to our Terms & Conditions and Privacy Policy.

Trellis went on to secure backing from CSIRO-founded VC Main Sequence and build speech recognition and translation software that works on networks disconnected from the internet. In Brisbane, it trialled crane-mounted cameras with the federal agriculture department on the docks to spot pests hitching a ride on shipping containers.

The launch comes as the Albanese government prepares to legislate mandatory safety standards for frontier AI companies, including independent model assessments and incident reporting.

Michael Gately, the CEO, said the company’s products already meet the standards the government outlined this week.

Début de l'article, extrait de Startup Daily. Pour lire la suite, ouvrez l'article chez le média.