Google's new Gemini 2.0 Flash model powers its namesake AI chatbot

Just a year after launching Gemini 1.0, Google is back with its successor. The Gemini 2.0 model is now here, and the Mountain View tech giant also launched Gemini 2.0 Flash, which built on the success of Gemini 1.5 Flash months ago.

Both models have their multimodal capabilities with text, images, and audio, but there’s a lot to say about Gemini 2.0. Google says that Gemini 2.0 Flash, albeit currently experimental, is now available on the Gemini chatbot on both desktop and mobile web. Soon, it will be available in the Gemini mobile app for iOS and Android.

But just how good is Gemini 2.0 in its application, though? Available globally through Google Search as well, Gemini 2.0 takes the cake from 1.5 in summarizing texts, analyzing images, transcribing long audio, debugging codes, storytelling, and even more. It’s a big leap, especially knowing how Gemini 1.5 had problems of its own in the past.

Google says that Gemini 2.0 Flash excels across several benchmarks compared to Gemini 1.5 models. It achieves higher scores in Natural2Code (92.9%), FACTS Grounding (83.6%), and MATH (89.7%), demonstrating a better performance in code generation, factual accuracy, and solving challenging math problems.

The now Gemini-powering model also shows progress in Bird-SQL (56.9%) and HiddenMath (63.0%), while slightly trailing in long-context understanding (69.2%) compared to Gemini 1.5 Pro. When it comes to reasoning, Gemini 2.0 Flash scored 62.1% on the GPQA (diamond) benchmark, which is a slightly big jump from 1.5 Flash.

Google has also made several announcements for the “agentic era” of AI, including projects like Project Mariner, which helps AI work within browsers to complete tasks, and Project Astra, which explores how AI can act as a real-world assistant. They also introduced Jules, an AI tool to help developers write code, and Genie 2, which creates 3D gaming worlds.