Every major tech company is now racing to make the most advanced AI in the world. Companies such as OpenAI, Anthropic, and Meta make sure that AI will be as powerful as possible. Started as a normal LLMs, and now, we are seeing a new generation of AI models that are multi-modal and agentic. Google is not late, they have released Gemini 2.0!
Google launched Gemini 2.0 just on Wednesday, December 11. The leaders behind Google AI are excited about the capabilities Gemini 2.0 could bring to the world.
As for performance, it is comparable to their previous top-tier AI, the Gemini 1.5 Pro. It seemed to be cost-efficient in terms of efficiency, and the speed is relative to current public models. What improved is its ability to do other things, which is doing tasks outside of chat.
Google is very pleased with what Gemini 2.0 turned out to be. Aside from being an LLM, it is now able to do tasks on the computer with the permission of a human.
Gemini 2.0 is reintroduced in Google’s app called Project Astra. For context, Project Astra was introduced this Spring of 2024. It is an app that enables Android and smart glasses to be very smart assistants capable of real-time conversation.
Think of it as Jarvis in Iron Man movies. It’s capable of analyzing your surroundings and helping you when you need assistance. It is possible through the use of camera and audio. Gemini 2.0 boosted Project Astra’s capabilities. Making it more like an AI agent and providing enhanced multimodal capabilities.
They also plan to release a real AI agent via the Chrome extension. In the Gemini 2.0 release, they introduce the Project Mariner. It is a brand-new Chrome extension that provides automation on the browser.
2025 is when the agent-based era takes off, as per Hassabis, and he’s putting Gemini 2.0 at the center of it. He’s quick to point out it’s not just about better performance. While there’s talk in the industry about AI improvements slowing down, Google is still making progress with new models. What’s got him just as hyped? The leaps in speed and efficiency.
Unsurprisingly, Google wants Gemini 2.0 everywhere. It’s set to power AI Overviews in Google Search and is already reaching a whopping 1 billion people. It’s only going to get more sophisticated as days pass by
You’ll see it in the Gemini bot and app, and soon it’ll drive AI features in Workspace and beyond. Google’s strategy? Pack as much functionality as possible into Gemini itself, so it can do more in more places. To put it simply, they’re building the most general model possible.
With agent-based AI, old and new challenges are on the table. The classic performance, efficiency, and cost are still here. But the new ones? They’re a bit of a wildcard.
Take safety risks: what happens when these agents operate independently in the real world? Google is being cautious, testing tools like Mariner and Astra and experimenting with “hardened sandboxes” to keep things contained. As Hassabis says, “Agents will be more useful, but also come with more risks.”
Gemini 2.0 might be experimental for now, but you can already check it out in the Gemini web app. Just pick the new model. As for the non-Flash versions? Those are coming early next year along with broader rollouts across Google platforms. Gemini 2.0 is just getting started and we could see more capable models in the upcoming years.