Before diving into the praise for Gemini, allow me to share two brief interludes: a holiday side project and Perplexity's new Lab feature.
During the three-day holiday, I spent time working with Claude-4 as planned:
- Using OpenAI's Deep Research to analyze key US-listed companies;
- Utilizing only the Claude-4-Sonnet model within Cursor to visualize and generate knowledge bases from the aforementioned Deep Research results;
Overall, this process was quite smooth. In this type of open-ended project generation work, the stability of Claude-4-Sonnet has significantly improved; compared to Claude-3.7-Sonnet, the projects are able to progress much further.
Below are screenshots of some key pages.
Of course, there are some notable drawbacks: although you can initiate searches directly in Cursor and have the model process them or handle text-based information, limited by the model's own context constraints, there are still many minor issues regarding information accuracy. This is an area where Cursor should improve in the future.
Actually, I've always believed that AI applications only need an IDE environment—after all, what task can't be accomplished with a piece of program code?
While "wandering around," I discovered that the Gemini app has added a new feature: yes, a 'Create' function has appeared in the top right corner of Deep Research results.
Then, remembering that Perplexity also added a 'Lab' feature a few days ago that I hadn't tested yet, I decided to check out Perplexity's 'Lab' first before returning to Gemini.
First, the 'Lab' feature adds a '' icon next to the original search and Deep Research options.
Of course, the icon isn't the point; the result is: I tried to have it complete an in-depth research piece on "Apple Inc."
The execution process is quite similar to a more aesthetically pleasing version of Manus.
Naturally, the smoothness and results are still better than Manus (completed in 5 minutes, though the quality of the result still lags significantly behind GPT and Gemini).
Summary: This is the smoothest and most complete one-click knowledge generation feature I've seen so far. Search, Python code, page code, direct display, and package downloads—it has everything. In fact, most AI features don't require "Browser Use" or "Computer Use," which only increase complexity, decrease execution speed, and easily introduce various errors.
However, even though I rate it highly, I still won't use this feature in my production workflow for a simple reason: the generation quality isn't good enough.
Of course, if Gemini didn't exist, this might suffice, but Gemini's new feature update has truly impressed me again (let's get to the main point).
Alright, after I performed the same "Apple Inc." Deep Research using Gemini-2.5-Pro, I suddenly noticed some additional options under the 'Create' function in the top right: recently, it only had "Audio Overview" for generating a podcast.
I clicked "Web Page" directly. Previously, I always moved Deep Research results into the Claude app for visualization; now, Gemini has integrated the "Page Generation" feature directly.
If it were just the aforementioned page, it would essentially be tied with Claude. Gemini, of course, doesn't stop there; it also added a "Generate" function. By clicking the "Generate" button in the bottom right, you can modify the webpage code and directly add "AI-generated" buttons.
For example, the page above can directly explain terms like "PE" (Price-to-Earnings ratio) and perform analysis.
Furthermore, it can add competitive strategy analysis when analyzing competitors.
The code clearly illustrates how to call APIs, and it even has the prompts written out.
Clearly, in the competition with ChatGPT and Claude, Gemini is becoming increasingly dominant.
While I don't necessarily believe Gemini's Deep Research surpasses ChatGPT, nor do I think Gemini-2.5's coding ability is definitely stronger than Claude's, the current Gemini application has integrated the entire workflow—from Deep Research to visualization and other productization processes.
In a context where capabilities are similar, the value of Claude and ChatGPT has suddenly diminished. Perhaps it's time to cancel those subscriptions.
Moreover, a better version of Gemini can be used for free in AI Studio.
Perhaps we will soon find that in the face of the best models, the space left for other AI applications to grow is indeed shrinking rapidly.