Google is testing functionality to upload documents in Gemini Deep Research

Google Gemini Deep Research might get a significant upgrade soon. Google is testing a new feature that allows users to upload files when using Google’s Deep Research tool.

Tech news blogger and creator of TestingCatalog Alexey Shabanov has studied the Gemini toolbox, also known as Agents, for upcoming new features.

He has found several new features that might be revealed at the Google I/O event that takes place on May 20 and 21.

According to the beta tester, Google Gemini’s Deep Research tool will most likely receive a significant upgrade. This new feature enables users to upload files, including images, source code, and documents, for more tailored research insights.

“The ability to ground research against user-uploaded files would make Deep Research much more versatile for professionals and researchers,” Shabanov explains. Thus far, the AI ​​tool only uses online sources.

The ability to load files into Gemini Deep Research is a highly requested feature and is most likely to roll out somewhere in the coming months.

Another new element that Shabanov found is the MMGEN Discovery Card. This appears to be a discovery pop-up for multi-modal generation. According to the beta tester, this aligns with plans to introduce a new subscription plan called Gemini Ultra, which consolidates advanced features for video and image generation under one plan.

In addition, the Gemini toolbox suggests that Google is expanding to include options like Memory (Teamfood), Veo 2 for video generation, and Image, alongside existing tools such as Canvas, Deep Research, and Search.

Shabanov’s conclusion is interesting. He’s suggesting that Google is trying to provide more granular subscription options for Gemini users, and to keep pace with competitors like OpenAI and Anthropic.

“These changes reflect Google’s ongoing shift toward subscription-based AI services, aiming to serve both casual users and professionals needing advanced capabilities,” the tech blogger concludes.