I tested the new Deep Research version and it’s surprisingly good

When Google added the Deep Research option to Gemini, I tested it right away. That was back in December, and my conclusion was that it was somewhat of a hit and miss. While the feature worked well for relatively basic inquiries, in-depth research on recent and slightly more complicated topics proved challenging for it.

Recently, Google announced the new and improved Deep Research, which is now based on the Gemini 2.0 Flash Thinking Experimental model. But how much has it improved, and can it now handle more complex research tasks? I tested it to find out.

If you’re already familiar with Deep Research, feel free to skip to the next section. Otherwise, read on.

Deep Research is designed to save you time when researching a specific topic. For example, let’s say you want to learn about all the components that go into a smartphone and their functions. This would typically require extensive online searching and reading multiple sources, likely leaving you with additional questions and continuing the cycle. It can be a very time-consuming and somewhat frustrating process, at least in my experience.

Deep Research aims to do that work for you. Tell the chatbot you’re interested in learning more about smartphone components and their functions, and voilà, Deep Research will generate a comprehensive report. It scans dozens of sources in minutes and presents the findings on one page.

The best part is that if you don’t understand something, you can ask the chatbot to simplify it. If you have additional questions, just ask away. So basically, instead of opening countless tabs and getting lost in your research, you can manage everything within a single interface much more quickly.

Well, at least that’s the theory. That’s how Deep Research should work, but in my initial testing, I encountered a few issues that I hope Google has now resolved.

The good news is that while Deep Research was initially available only to Gemini Advanced users, the latest update makes it available for free to everyone. So if you have a Google account, you can try out the feature for yourself.

Research is something I enjoy, so I was excited to see how much better Deep Research has become. To test this properly, I’ll use the same prompts I used during my first review of the feature. This will provide a clear comparison and show how much it has — or hasn’t — improved.

The first test I ran was regarding the stock market, which I enjoy following as a finance enthusiast. Deep Research should technically be a great tool for providing a summary of a specific trading day, potentially saving a lot of time. Researching via Google Search, for instance, often requires visiting multiple websites to gather information, and even then, some questions might remain unanswered.

My first attempt with Deep Research yielded unimpressive results, making this an excellent test to see the tool’s improvement. This is the prompt I used:

I’m happy to report a significant difference between the first version of Deep Research and the latest one. The new version is much faster, providing the report in 2 minutes and 44 seconds, compared to the four and a half minutes required by the original version.

Interestingly, it used fewer sources (46 vs. 73) but categorized them into those used for the report and those consulted but ultimately set aside. Despite using fewer sources, the end result was vastly superior; the provided report was almost exactly what I requested.

This time, there were no glaring errors. For example, I asked for the performance of the top 10 S&P 500 companies by market cap, and the report included the correct information. When I first tested Deep Research in December with the same request, the report included only 9 instead of 10 companies, none of which were among the 10 biggest in the US. Furthermore, percentage changes were missing for four of those listed, rendering the information useless.

So, Deep Research performed substantially better on this task, although it wasn’t perfect. While it correctly identified the top-performing stocks in the S&P 500, it omitted some of the underperforming ones. Therefore, while not flawless, the report offered significant value, unlike the useless one generated by the first version.

Moving on to my second test, I asked Gemini to research different styles of interior design and color schemes and which ones are best suited for a small apartment in a modern building. I need to furnish my new apartment, so this is a topic I’ve been very interested in lately.

This task should theoretically be simpler for Gemini than the stock market analysis. It performed well during my initial testing, but the new version of Deep Research takes things to a new level. The report I received was much more detailed this time and included a wealth of valuable information. However, one major issue remains: the lack of images.

A report discussing interior styles should ideally include images, but it currently does not. When I asked Gemini to provide images, it stated it couldn’t display them in this format, which is unfortunate. Some reports are fine with text only, but others, like this one, would be greatly enhanced by images. Otherwise, I have to search for visuals myself using Google Search or Pinterest. The core idea behind Deep Research is to consolidate information within one interface, eliminating the need to manually visit multiple sources. Hopefully, this feature will roll out soon, although there’s no guarantee.

For my third test, I asked Gemini to research the topic of AI chatbots and detail which ones exist, their capabilities and limitations, their costs, and the future outlook for them. Deep Research performed better than the first time, providing a more in-depth look at the topic. The report takes longer to read, but it offers more substantial value.

It provided detailed information on chatbot types, major market players, their capabilities, limitations, and other relevant points. Generally, I was very satisfied with the report provided.

Based on my testing, the new and improved Deep Research feature is very effective. It’s much better than the original model released in December. While the original version struggled with specific research tasks — like the stock market overview — the new one handled them a lot better.

Sure, it isn’t perfect, and there might be other limitations I haven’t encountered, but based on my hands-on experience, it’s a tool many people will find beneficial. Whether you’re doing research for school or a personal project, I’d advise you to try it out and see how well it works for you. It’s free, so there’s nothing to lose.

It’s remarkable how much Deep Research has advanced in just a few months. AI is rapidly evolving, so one can only imagine how much better Gemini and features like Deep Research will become in a year or two.

Have you tried the improved version of Deep Research yet? Share your experience in the comments section.