.
Image: Supplied
Google’s newest large-scale AI model, Gemini 3, isn’t just another incremental upgrade, it’s a statement of intent.
The company is betting big on agentic AI, a next-generation paradigm where models don’t just respond to tasks, but plan, reason, and act with near-autonomy.
And in South Africa, where AI-skill searches have nearly doubled in the past year, Google sees an opportunity to accelerate a continental shift.
“For South Africa, the launch of Gemini 3 is more than a technological breakthrough, it is about democratising access to the most advanced intelligence,” says Kabelo Makwane, Google SA’s country director.
He frames the launch as an economic catalyst as much as a technical one: a model aimed at real job creation, skills development, and local relevance.
At the center of the announcement is Gemini 3 Pro, a model Google describes as capable of “PhD-level reasoning.” It sets new records on several AI benchmarks, including a standout 91.9% on the GPQA Diamond test, one of the industry’s most challenging measurements of complex reasoning.
But Google is equally focused on the texture of the model’s output. Gemini 3 promises more direct, less meandering responses, what the company calls “smart and direct” discourse. Developers get a new superpower: “vibe coding,” the ability to describe a feeling or concept and have the model generate software aligned to it.
The model can also plan and execute multi-step tasks, making it more of a collaborator than a tool.
For the first time, Google is launching a new Gemini model directly inside Search on day one. With “AI Mode,” Search becomes less like a query box and more like a dynamic learning environment.
Ask how a three-body physics system works and Search might generate an interactive simulation. Ask about mortgages and it builds a bespoke loan comparison calculator.
The shift signals Google’s vision for Search as not just a gateway to information, but a canvas for computation and exploration.
Inside the Gemini app, the company is rolling out a new agent that can act on your behalf across Google’s ecosystem.
For Google AI Ultra subscribers, this experimental agent can triage inboxes, draft replies, reference emails for travel bookings, compare rental cars, and prep reservations.
Google stresses that users remain “in control,” but the intent is unmistakable: everyday workflows are becoming co-managed by autonomous systems.
Perhaps the most forward-looking piece of the announcement is Google Antigravity, a developer platform designed for building agentic apps. Antigravity functions like a mission-control environment that orchestrates multiple agents, giving developers oversight through verifiable artifacts while letting the system autonomously plan and complete software tasks—from debugging to feature development.
If it works as promised, Antigravity could mark one of the first mainstream steps toward autonomous software engineering.
Google says Gemini 3 has passed its most extensive safety evaluations to date, a response to an industry increasingly under scrutiny for hallucinations, bias, and overconfidence.
Gemini 3 Pro is available through the Gemini app, and in Search via AI Mode for Google AI Pro and Ultra subscribers.
FAST COMPANY (SA)