Fact Checked by 
Google has released the first version of Gemini 2.0, its next-generation large language model, and the focus is on AI agents that do work on your behalf.
Gemini 2.0 Flash has upgrades that make it more useful for agentic AI, including multimodal output. It can now produce natively-created images alongside text, or make “steerable” text-to-speech audio in multiple languages. The LLM can also directly access Google searches, run code, and perform third-party user-defined tasks.
The move allows for agents that can not only perform multiple steps, but work across domains and better understand physical objects. Accordingly, an update to Project Astra with the new model can use Google’s search, Lens, and Maps to better interpret what you see. It can speak in multiple and even mixed languages, and remembers more of your recent and past conversations.
Google is also hinting at its long-term plans with new prototype AI agents. Project Mariner uses Gemini 2.0 to explore “human-agent interaction,” currently in the web browser. It can navigate and perform tasks on the web by understanding and reasoning around what’s on screen.
In Gemini Advanced using the 1.5 Pro model, Deep Research is an early AI assistant that can create a multi-step research plan and analyze info once you’ve approved the strategy. It doesn’t just search the web, Google notes — it “refines” what it learns through multiple searches, and creates a report with linked sources. This theoretically finishes research in minutes that would normally take hours.
Gemini 2.0 and Deep Research availability
Google is making Gemini 2.0 Flash available now as an experimental model. Worldwide, you can use a “chat optimized” version on the desktop and mobile web. The native Gemini mobile app will have access in the near future, and 2.0 will come to more Google services in 2025.
Project Astra will soon be available for a “small group” of testers using prototype smart glasses. Project Mariner is already available for trusted testers through a Chrome extension, but Google wants to talk to web creators to make sure any wider work is deployed “safely and responsibly.”
Deep Research is available today for Gemini Advanced subscribers in English on desktop and mobile browsers, and comes to mobile in 2025.
There’s pressure to roll out 2.0 relatively quickly. Many companies see AI agents as the eventual replacement for apps, as they can potentially handle many duties at the same time. Former Google and Meta leaders recently formed a startup, /dev/agents, devoted to making a platform for agentic AI. The Gemini revamp could be key to maintaining Google’s clout in generative AI as the technology evolves from its narrow-purpose origins.
