Airtree backs Adelaide startup after its robotics model beats Alibaba and Mistral
Metacognition raised a massive pre-seed round to develop an AI operating system that helps robots navigate the physical world safely.
The short version
- Metacognition raised $15 million in a pre-seed round, nearly 10 times the 2026 median for that stage.
- The startup released NavGPT-3, an open-source model that helps robots interpret verbal commands and navigate.
- Simulated tests showed the model outperformed navigation tech from Alibaba and Mistral.
- The company aims to build an 'operating system' for robots to ensure they act safely in homes and industries.
Why it matters
This represents a shift toward Australia producing foundational AI software rather than just hosting data centers. It also highlights the growing need for 'guardrails' in robotics to prevent physical accidents caused by AI errors.
A massive pre-seed investment
Airtree, one of Australia's largest venture capital firms, has invested $5 million into the Adelaide-based startup Metacognition. This injection of capital brings the company's pre-seed funding total to $15 million, following an initial $10 million investment from Main Sequence.
While $15 million is not a massive sum in the broader tech landscape, it is considered astronomical for a pre-seed round. This stage of funding is almost as early as 'early-stage' investing, and the amount is nearly 10 times the 2026 median of $1.6 million.
Co-founder Anton van den Hengel describes the current funding as 'walking around money' compared to what the company will eventually need. To compete globally, the startup plans to grow from its current team of 20 people to a workforce of 100, most of whom will hold PhDs.
Beating global tech giants
Metacognition recently made headlines by releasing NavGPT-3, an open-source vision-language model. The technology is designed to allow robots to interpret verbal commands and navigate through various environments by combining two types of AI: reasoning and navigation.
In simulated benchmarking tests, the model demonstrated more reliable navigation than competing technology from the $360 billion Chinese giant Alibaba and the $24 billion French startup Mistral. These results were particularly notable because they were achieved using the model on its own, even though the strongest results came from integrating frontier models from OpenAI and Anthropic.
The success of NavGPT-3 serves as a proof of concept for the Australian tech scene. Van den Hengel stated that Australia can do foundational AI and turn cutting-edge AI into economic value, highlighting the country's potential to move beyond being a mere host for data centers.
The robotics operating system
The company's core ambition is to develop an 'operating system' for robots. This would function as an AI layer that allows robots to understand human commands while establishing specific guardrails to prevent unsafe or counterproductive actions in the physical world.
The developers argue that while AI hallucinations in software like ChatGPT are problematic, the stakes are much higher for robots. Because robots interact with the physical environment, they have the potential to cause actual harm if they do not have a reliable way to interpret instructions safely.
Metacognition hopes to become the standard software provider for robotics manufacturers worldwide. The goal is to create a platform similar to how many computer makers rely on Microsoft's Windows to run their hardware.
Economic value and ownership
The investment from Airtree reflects a desire to ensure Australia participates in the core of the AI process. Airtree co-founder Craig Blair noted that the country should not just 'keep the lights on' for models trained in other countries, as that allows the primary value to leave the country.
By backing the people building the 'missing pieces' of the AI stack, investors hope to capture more economic value locally. This strategy focuses on creating foundational technology that can be exported and utilized by various industries rather than just providing infrastructure for foreign companies.
Van den Hengel expressed a desire for Australia to participate in the core of the AI process rather than just being a real estate transaction for someone else's growth. The startup's success suggests that Australia can compete in high-level AI development. By focusing on the specific intersection of reasoning and physical navigation, Metacognition is attempting to carve out a niche in the global robotics market.
Strategic investment goals
Airtree is casting its vote in a specific direction for the Australian tech landscape. By backing Metacognition, the venture capital firm is supporting a move toward domestic participation in the core of the AI process rather than just serving as a real estate transaction for external growth.
While the startup is making waves, the firm acknowledges that in absolute terms, there is little notable about an AI startup raising $15 million. However, the strategic importance lies in the focus on building the missing pieces of the technology stack to ensure local economic value.
What to watch
- How Metacognition scales its team from 20 to 100 people.
- Whether other Australian firms follow suit in seeking 'pre-seed' funding at these higher levels.
- The adoption rate of the 'robotics operating system' by global hardware manufacturers.
Try it
Look up the open-source NavGPT-3 model to see how it handles vision-language tasks compared to other navigation models.
Sources and links
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by Forbes Australia · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
