Google announced Gemini 2.0, their most capable AI model yet, designed for what they're calling "the agentic era." And if you're wondering what "agentic" actually means in practice, you're not alone.
The New Model Is Fast (and Free)
Gemini 2.0 Flash is available to developers and trusted testers, with wider availability planned for early next year, and Gemini users globally can try out a chat optimized version by selecting it in the model dropdown on desktop. The "Flash" designation means it's their faster, lighter model—similar to how GPT-4o is faster than regular GPT-4.
Gemini 2.0 Flash even outperforms the previous 1.5 Pro on key benchmarks, at twice the speed. So Google basically made their cheaper, faster model better than their premium model from six months ago. That's the pace we're moving at now.
I tried the Flash 2.0 yesterday through the web interface, and yeah, it's noticeably snappier. Responses come back faster, and for basic queries it feels just as capable as the slower models. Whether that holds up for complex tasks is another question.
Agents, Agents Everywhere
Here's where Google's really pushing: agents. Over the last year, Google has been investing in developing more agentic models, meaning they can understand more about the world around you, think multiple steps ahead, and take action on your behalf, with your supervision.
They demoed Project Astra (a universal AI assistant), Project Mariner (which can navigate Chrome and click things for you), and some other prototypes. The idea is that instead of just answering questions, these models can actually do stuff—book appointments, fill out forms, research complex topics autonomously.
Someone I know who got early access to Project Mariner said it's simultaneously impressive and deeply unsettling. Watching an AI take control of your browser and navigate websites is cool until you realize how many things could go wrong. What happens when it clicks "purchase" by mistake? What if it misinterprets an instruction?
Multimodal Is the Real Story
Gemini 2.0 now supports multimodal output like natively generated images mixed with text and steerable text-to-speech multilingual audio, and it can natively call tools like Google Search, code execution, and third-party user-defined functions.
This is the part that actually matters for most users. Previous models could understand images and video. Now Gemini 2.0 can generate them too, plus audio, all in the same response. Ask it to explain a concept and it might give you text, a diagram, and a spoken explanation all at once.
The text-to-speech is particularly interesting. "Steerable" means you can control how it sounds—different accents, different speaking styles, different levels of formality. That's useful for accessibility, but also kind of weird when you realize AI voices are about to get really good at sounding however they want.
How This Compares to the Competition
Gemini 2.0 represents Google's latest efforts in the tech industry's increasingly competitive AI race, competing against Microsoft, Meta, OpenAI, and Anthropic. Everyone's releasing new models right now. OpenAI has ChatGPT Search. Anthropic has Claude 3.5 Sonnet. Microsoft has... Copilot everything.
Google's advantage is integration. Gemini powers all 7 of Google's products with 2 billion users, including Search where AI Overviews now reach 1 billion people. When your AI model is baked into Gmail, Docs, Search, and everything else, distribution isn't a problem. Adoption is.
The question is whether people actually want AI deeply integrated into these products, or if they prefer AI as a separate tool they use when needed. I use Google products daily and honestly haven't incorporated the AI features much. The suggestions are sometimes helpful, sometimes annoying, rarely transformative.
Deep Research Is Legitimately Cool
Google launched a new feature called Deep Research, which uses advanced reasoning and long context capabilities to act as a research assistant, exploring complex topics and compiling reports on your behalf, available in Gemini Advanced today.
I tested this on a topic I knew well (AI regulation timelines across different countries) and the report it generated was... actually pretty good? It cited sources, organized information logically, and caught some developments I'd missed. Not perfect, but way better than I expected.
The catch: it's only available to paid Gemini Advanced subscribers. Google's clearly positioning this as a premium feature. Makes sense—running these long research sessions is expensive.
The Agentic Future (Allegedly)
If Gemini 1.0 was about organizing and understanding information, Gemini 2.0 is about making it much more useful. That's Google's pitch. Models that don't just answer questions but actively help you accomplish tasks.
Will it work? Depends on whether people actually trust AI agents to do things on their behalf. Right now, I don't. Maybe in a year or two that changes as the models get more reliable and the error rates drop. But we're still in the phase where AI confidently gives wrong answers to basic math problems.
The agentic vision requires a level of reliability we haven't achieved yet. Answering questions wrong is annoying. Taking actions wrong is potentially expensive or harmful. The stakes are different, and I'm not convinced the technology is ready.
My Take
Gemini 2.0 is a solid incremental improvement. Faster, more capable, more integrated. The agent stuff is interesting but feels early. Deep Research is genuinely useful for a specific use case.
Is it a game-changer? No. Is it keeping Google competitive in a crowded market? Absolutely. And given that this is their "Flash" model—supposedly the budget option—I'm curious what Gemini 2.0 Pro will look like when it eventually ships.
For now, it's free to try if you're curious. Just don't expect it to revolutionize your workflow overnight.