Powrót do Wiadomości
ProduktAI Understanding odprawa

Trellis Data launches 220B-parameter sovereign AI model Agentaus 1

Australian firm Trellis Data has released Agentaus 1, a near-frontier 220 billion-parameter model it says ranks 43rd among foundation models, targeting local business and government customers first.

4 min readRead the linked source
Source-provided image accompanying Trellis Data launches 220B-parameter sovereign AI model Agentaus 1
Odniesienie do źródłaŹródło zapisane
Wydawca
technologydecisions.com.au
Typ źródła
Źródło powiązane — nie ustalono statusu źródła pierwotnego.
KontekstZrozum to w 60 sekund

Kluczowe terminy

Parametr
Wyuczona waga wewnątrz modelu, która wpływa na jego wyniki.
API (interfejs programowania aplikacji)
Ustrukturyzowany sposób wysyłania żądań przez jeden system oprogramowania i otrzymywania odpowiedzi z innego systemu.
Model dużego języka (LLM)
Model językowy wyszkolony na ogromnych korpusach tekstowych w celu generowania i analizowania tekstu.

Co się stało

Australian AI company Trellis Data has launched Agentaus 1, a 220 billion- large language model that Technology Decisions describes as one of the most advanced AI models developed in Australia to date. The company says the model is roughly 50% larger than its predecessor and claims a world ranking of 43 for foundation models, though the outlet does not specify who produces that ranking or how it was measured. Trellis Data says Agentaus 1 is designed for greater accuracy, deeper multi-step reasoning and more coherent outputs, while remaining secure enough for sovereign workloads and efficient enough to run at scale on smaller compute clusters. The company plans to target Australian businesses first before expanding internationally, with access offered through free, subscription, enterprise and API tiers.

Technology Decisions reports that Trellis Data, an Australian AI company that has been building AI capabilities for high-security environments for a decade, has launched Agentaus 1, a near-frontier large language model with 220 billion parameters. The outlet describes it as one of the most advanced AI models developed in the nation to date.

According to the report, the model is around 50% larger than Trellis Data's previous model and has achieved a world ranking of 43 for foundation models. The outlet does not name the source of that ranking or provide benchmark details, so the figure cannot be independently confirmed from this report.

The company says Agentaus 1 was designed to deliver greater accuracy, deeper multi-step reasoning and more coherent outputs while remaining secure enough for sovereign workloads and efficient enough to operate at scale on smaller compute clusters. Founder Michael Gately said the model gives Australian organisations a more secure, effective and cost-efficient option while keeping their data and investment in Australia.

Trellis Data plans to target Australian businesses first before expanding into international markets. The report states that access is available via free, subscription, enterprise and API tiers, but no pricing, availability dates or technical specifications for those tiers are documented in the source.

Szczegóły źródła: technologydecisions.com.au

Dlaczego to ma znaczenie

The launch is a concrete attempt to give Australian organisations a domestically developed alternative to models from overseas AI giants, which founder Michael Gately frames as a matter of productivity, economic prosperity and national resilience. If the company's performance and security claims hold up, government and high-security customers could run advanced AI workloads without sending data or investment offshore — a positioning that aligns with the Australian Government's planned mandatory AI safety and security protections for frontier AI, which Gately says Trellis Data products already meet. The claim that the model can operate on smaller compute clusters also matters practically, since efficiency at that scale would lower the infrastructure barrier for local adoption. However, the ranking claim, benchmark results and compliance assertions are the company's own statements relayed by the outlet and have not been independently verified.

The launch is positioned as a sovereign AI play: Australian business and government customers could reduce reliance on overseas models and avoid accessing multiple AI models for different tasks, according to the company's statements in the report.

Gately links the release to the Australian Government's planned mandatory AI safety and security protections for frontier AI, claiming Agentaus 1 and all Trellis Data products already meet these standards. The government regime is not yet final, so this compliance claim is a company assertion rather than a verified certification.

The claimed ability to run at scale on smaller compute clusters is practically significant if confirmed, because it would lower the hardware cost of adopting a large model for organisations that cannot access frontier-scale infrastructure.

The performance ranking, security properties and efficiency claims all come from the company via Technology Decisions. No independent testing, third-party benchmark or customer deployment evidence appears in the report, and these claims should be treated as unverified.

Interactive Mechanism

Mechanizm interaktywny: jak to faktycznie działa

Poznaj interaktywnie technologię leżącą u podstaw tego rozwoju.

Thinking Budget (Test-Time Tokens):1,024 tokens
Complex Accuracy79%Math & Code Logic
Latency3.2sTime to first full output
Inference Cost$0.0092Per query estimated
Reasoning StyleStep VerificationInternal chain depth
Active Thinking Trace:
1Deconstruct user problem into formal constraints
2Propose candidate hypotheses & step-by-step calculation
3Self-correction: Backtrack and refute subtle edge cases
4Exhaustive consistency check & final output synthesis
Core takeaway: Test-time compute fundamentally changes AI economics. Instead of only scaling during pre-training, giving reasoning models more tokens at inference time allows them to systematically solve PhD-level STEM problems.
Interaktywna kontrola koncepcji+10 Points
What is AI? Quiz

A route planner searches possible journeys using explicit rules. What does this illustrate about AI?

Co obejrzeć dalej

Watch for independent benchmark evaluations of Agentaus 1, since the world ranking of 43 is a company claim without a named third-party evaluator in the report. Pricing details for the subscription, enterprise and API tiers are not documented and will determine how accessible the model actually is. The Australian Government's mandatory AI safety regime is still in development, so whether Trellis Data's products formally meet those standards remains to be seen. The company says it is already working on what comes next, and its planned expansion into international markets will test whether a sovereign Australian model can compete beyond its home market.

Independent benchmark results for Agentaus 1, which would substantiate or challenge the claimed world ranking of 43 and the near-frontier designation.

Pricing and concrete availability details for the free, subscription, enterprise and API tiers, none of which are documented in the report.

The final form of the Australian Government's mandatory AI safety and security protections, and whether Trellis Data's products are formally assessed against them.

Evidence of actual adoption by Australian government or business customers, and the timing and reception of the company's planned international expansion.

Powiązane przewodniki i quizy

Uznałeś to za przydatne?