Google Unveils Gemini 2 0 And Deep Research For Advanced Users

Not only that, it supports native multimodal output such as native image generation and native text-to-speech multilingual audio output. It can also natively interact with Google Search and perform code execution as well. As for benchmarks, Gemini 2.0 Flash achieves 62.1% in GPQA (Diamond), 76.4% in MMLU-Pro, and 92.9% in Natural2Code.
Gemini 2.0 Flash is now available on the web version of Gemini to all users including free and paid users. You need to click on the drop-down menu and select the “Gemini 2.0 Flash Experimental” model to use it. Google says the Gemini 2.0 Flash model will soon be added to the Gemini app on Android and iOS.
As for paid Gemini Advanced users, they get access to a new feature called “Deep Research” which uses advanced reasoning to solve complex queries. It can also help you compile reports for you. It seems Gemini Advanced users have access to something like OpenAI’s o1 reasoning models that can “think” step-by-step and reason through harder questions.
Google says Gemini 2.0 is coming to AI Overviews early next year. AI Overviews will be able to understand complex queries, advanced math equations, coding questions, and multimodal input. Notably, the search giant says the Gemini 2.0 model was trained entirely on its custom TPUs like Trillium, and inference is also done on its in-house TPUs.