Google today released a new option on Gemini called Deep Research. This feature is powered by Gemini 1.5 Pro and allows Gemini to perform tasks with reasoning through a chain of thoughts. It works slightly differently compared to other systems because, before execution, it first provides a research plan and prints it for user confirmation. The plan can also be edited, enabling more efficient research without waiting for improperly planned responses.
Introducing Deep Research, your personal agentic AI research assistant. Rolling out starting today in Gemini Advanced.
— Google Gemini App (@GeminiApp) December 11, 2024
With Deep Research, you can create in-depth research reports on complex topics, complete with source links, giving you hours of research at your fingertips in… pic.twitter.com/YMO466Ni6g
Once the plan is approved, Gemini starts the research and opens a canvas side sheet along with the chat. In this side-sheet, it shows the progress as a list of links and URLs Gemini has searched through. This is a significant addition since, for example, OpenAI’s reasoning mode doesn’t yet include search capabilities. The only comparable tool previously was Perplexity Pro with reasoning. Now, Gemini Deep Research acts as a competitor.
The key difference between Perplexity and Gemini Deep Research lies in speed. While Perplexity is optimized for quick responses, Gemini takes more time, reasoning that this can produce better results by searching across a broader range of resources. Gemini can process more than 50 links before responding, though the process typically takes over 2 minutes.
In my prompts, I asked it to visit the Testing Catalog News Blog, search for articles from the past month, categorize them, and print the count. While it didn’t complete the task perfectly, it came close in some attempts due to the extended time spent researching more pages.
This feature is currently available only to Gemini advanced subscribers and isn’t accessible in every country yet, though it works well with a VPN.
