Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Phoebe encoding It can be a lot of fun when you have the right mindset. It doesn’t take an engineer to have a good idea that could turn into something great, and that’s what makes building apps using only natural language so attractive — anyone can master it. I engaged in a fair amount of programming myself by doing this Event calendars And replay Childhood video games In my web browser, just by chatting with A chatbotBut I know I’ve barely scratched the surface.
The model you use can have a huge impact on the quality of a project’s output and I have witnessed this first-hand. I wanted to see how lighter models compare to “thinking” models, as Google and OpenAI refer to them. These lighter models differ in name: Google’s Gemini interface calls it Fast (although the model is actually called, for example, Gemini 2.5 Flash), while OpenAI It’s called a moment.
To get a sense of how each paradigm in dynamic programming differs, I conducted an experiment. First, I started by creating a project using the Gemini Thinking Model – Gemini 3 Pro – and wanted to see if I could replicate the same project with the Quick Model using the same prompts from the previous project. Since there was no way to guarantee answers for every form, I knew there would be differences and conversations would branch off, but for the most part, I tried to keep the conversations identical when I could.
At the time of this test, the fast model was the Gemini 2.5 Flash. I expected the end results to be different, and they were, but not nearly as much as I expected. What was different was how I went from A to Z with each model.
I was lacking inspiration for this experiment, so I just offloaded it to Gemini. I asked him to come up with interesting programming projects that I could work on and I chose one called “Trophy Display Case”. I asked Jiminy to display a list of horror movies, instead of awards, and to provide more information about them when I clicked on a poster. Beyond these requirements, both Gemini models were given creative control.
Don’t miss any of our unbiased technical content and lab reviews. Add CNET As Google’s preferred source.
Fast AI Models vs. Thinking Models: What’s the Difference?
If Google gives us a choice between the Flash and Pro models, they should be vastly different, right? Yes and no. They are both major linguistic models, but they work differently. For the everyday user, “Fast” and “Thinking” define the differences between the two well enough: speed vs. depth.
A Logic model It is an LLM finely tuned to break down complex problems into smaller steps before creating the final output. This is done by conducting an internal series of logical thinking paths. Both the Gemini 2.5 Flash and Gemini 3 Pro are logical models, but the Gemini 2.5 Flash takes Hybrid approach: It is presented by A Balancing act Between speed and logic.
Gemini 3 Pro is the most powerful thinking model, optimized for deep diving to find answers. As a result, it’s slower than more efficient models like 2.5 Flash. Google has since released it Gemini 3 flasha more powerful base model that replaced the 2.5 Flash. The Gemini 3 Pro remains the most powerful Gemini thinking model available for most people.
The Gemini 3 Pro model did most of the work
The final project created by the Gemini 3 Pro was not perfect, but it was better than my original idea and was about a mile ahead of what the Gemini 2.5 Flash produced.
Using the Gemini 3 Pro, I was able to create a landing page that showcased the movies from my list, complete with poster images, and when you clicked on the title, the page would open and reveal additional information, as well as a link to view the trailer on YouTube. The project wasn’t complicated, but I encountered a lot of issues and bugs along the way.
I originally wanted to embed the trailers on the page, but it kept reporting errors that Gemini couldn’t fix, so I scaled it back by only providing a linked image for watching the trailer on YouTube. It was good, but a less smooth experience than I wanted. However, I appreciated the way Gemini 3 Pro detailed the specific issues it was having with this feature and allowed me to make the decision to cancel it.
Another issue that the Gemini 3 Pro has tried to fix several times is what it describes as a layering issue. When you clicked on the poster, a pop-up window with movie details would be displayed, as well as a small button to exit that view, although it never worked. I asked Gemini to fix the problem four times, and it failed to fix the problem until that last request. Gemini explained what they were doing with the code in general, but didn’t go into a lot of detail, although I imagine they would have provided details if you’d asked.
The original project was just a way to showcase a collection of films and get more information about them. Outside of that, I didn’t think anything about design or ways to make a web app interesting, and the Gemini 3 Pro was helpful in that area. When I asked how I could improve the app, both in terms of design and features, he suggested adding a 3D wheel effect to movies and a random selection option.
This project took approximately 20 iterations. The final product was as good as it could get, and it was a fun project, but there were problems that Gemini often failed to fix. The final product did more than I expected, so I was happy with it. But with all the problems I encountered, I started to wonder how the Gemini rapid model would handle the same project.
Programming the Vibe with Gemini 2.5 Flash is more manual
Unsurprisingly, using the “Quick” model was faster than the Gemini 3 Pro, but for the most part, this model suggested more manual approaches to coming up with solutions for the project. The AI worked quickly, but it created more—and slower—work for me.
For example, I wanted the web app to display the poster and synopsis for every movie in the list, but I never thought about how to create this information. Without me specifically asking, Gemini 3 Pro suggested I could sign up for it Movie database And get an API key to pull these details automatically, as Gemini 2.5 Flash basically tells me to “grab” the images and go from there. How I got those photos seemed to be up to me.
The Gemini 2.5 Flash sometimes feels almost lazy compared to the Gemini 3 Pro. There are some things the Gemini Pro model will do without you being asked, but the Flash needs a more specific prompt. At times, I felt like I was prodding a child who heard instructions but deliberately avoided his chores.
In multiple cases, after I asked Gemini 2.5 Flash to make a change, it did so and provided the updated code, but only for the specific section that was modified. Then it will ask me to replace the old code with the new one. If you know what you’re looking at, replacing one section of code with another probably isn’t a big deal, but that’s dynamic coding, and if you don’t know where to put the code, even if it’s a really easy task, that might stop some people in their tracks. She can clear the atmosphere.
Furthermore, Gemini 2.5 Flash simply suggested that I “get” the movie poster images and additional details. So, while breaking the bounds of the experiment parameters to only use the same prompts in both projects (which were loose, at best), I decided to ask Gemini 2.5 Flash what it thought about the idea of adding an API key to the movie database. He welcomed the idea and told me where to add the key. Instead, I asked the form to add the key I provided. He added the switch, but when I ran the web app, it wasn’t actually pulling poster images from the movies I listed, so I had to ask him to fix that again. The model pointed out its limitations by saying, “It takes a long time to find the exact TMDB ID for every movie in your original list, but I would fill the array with as many confirmed IDs as possible to make the collection accurate to the requested list.”
If she did anything she said she would do, I didn’t notice. Compared to the wide range of different films that have appeared, any match from the list I’ve provided seems like a complete coincidence. However, even though 99% of the movie posters posted were incorrect, this theoretically prevented me from having to manually add the images themselves. In contrast, the Gemini 3 Pro filled in all the correct movie posters in one shot.
How did speed and thinking models differ in biological programming?
Every time I ask Gemini’s thought model to make an edit, it makes the change and rewrites all the code instantly so I can simply copy and paste the entire code anywhere I want without having to know where to update the code.
Gemini 2.5 Flash was different. At some point, after making a small modification, he gave me the code and asked me to replace it with what was there. Hoping to avoid this, I asked him to rewrite the entire code so I wouldn’t have to change anything. Her response: “That’s a huge ask.” Although he thought I was asking for more from it at that moment, it was a bit jarring when compared to the Gemini 3 Pro.
Projects
Gemini 2.5 Flash provided a fairly working project, but it was riddled with bugs even after trying to debug it.
At the end of testing, neither model was perfect, but working with the Gemini 3 Pro was much easier. While both models could, in another project, lead to very similar results, getting to that final destination will likely take two very different paths.
Working with Gemini 2.5 Flash, you need to be specific about what you want it to do and be prepared to correct it when it seems to take shortcuts. It will take practice and experience, including working with other AI models, to identify when a model takes a shortcut that could impact the project. If this is the only model you’re working with, you’ll need to be more diligent with it overall.
Gemini 3 Pro really deserves its name. Not only did he handle the heavy lifting of this project, but he also provided helpful suggestions that lifted it from the basic idea I started with.