You hold the weights
If you leave, they come with you.
Independent AI, built in Australia
We make Matilda and run it on GPUs we own in Melbourne, under Australian law and with no foreign parent company.
Maincode exists to build true technology independence for Australia. AI came first, with Matilda, and our own cloud comes next.
Dave Lemphers · Australian-made AI is independent AI
By the numbers
Every number links to its proof.
MC-2 throughput
~10ktok/s
Qwen3.6 35B on one MC-2 node, 10,000-token prompts, 25 May 2026
MC-2 benchmarks
Per-token all-reduce
1.3×
Faster than RCCL’s best on small-message all-reduce, 8 MI355X GPUs
The engine
Cost per token
87%
Lower at best than public APIs, 8,000-token prompts, owned MI355X
The economics
Maincoder-1B · HumanEval
76%
Highest of the 1B-class open models in our Dec 2025 comparison
Maincoder-1B
Matilda
Matilda shows each step and asks before it acts.
Chat, write code or hand off a whole task, on the web, your phone, the desktop app or a terminal.
Free to start, from $25 a month for Chat and Code. Compare plans
Matilda Code in your terminal, the desktop app and CI
Enterprise
We train a model you own and run it in your instance.
It trains on your documents and your people’s corrections, so what they know becomes your IP instead of another company’s training data.
Your own instance
It runs in Melbourne or in your own data centre, and no other customer’s model trains on your data.
Under Australian law
The contract keeps your data and IP in Australia, under Australian law.
You approve every version
Each new version trains on your people’s feedback and waits for your sign-off before it goes live.
Infrastructure
We run every layer, from the GPUs to the model.
MC-2 is the AMD MI355X cluster we own in Melbourne, and Matilda runs on it through an inference engine we wrote in Rust.
Inside our infrastructure
ProductMatildaChat, Code, Desktop and the API
PlatformFourteen steps around the modelSign-in, safety checks, search over your documents, audit trail
ModelsOpen weights, trained on your materialAligned and trained with the Matilda training system
RuntimeOur own inference engineWritten in Rust for the MI355X
HardwareMC-2, MelbourneBare-metal AMD MI355X GPUs, with MC-3 to follow
Latest from the lab
Recent posts on our research and engineering.
Get in touch